Beyond Vector Search: Building a "Reasoning Engine" in Copilot Studio

Intro:
We've all been there. You build a powerful RAG copilot, feed it a dense, 100-page document, and ask a specific, nuanced question... only to get a vague, incomplete, or just plain wrong answer. Why does this happen?

The culprit is often our reliance on traditional vector search, which excels at finding "semantically similar" text but struggles to understand context, nuance, and structure. It finds words that sound like the answer, but it can't reason about the document to find the truth.

One of my work colleage shared about PageIndex which made me curious to explore.

The Experimental Setup:

To create a fair and direct comparison, I used a single Microsoft Copilot Studio instance and one source document: the official EU AI ACT .pdf. From this document, I generated two distinct knowledge sources. The first was a standard embeddings.json file, created by chunking the document and generating vector embeddings—the foundation for a traditional RAG approach. The second was a structured document-structure-pi.json file, which acts as a hierarchical "Table of Contents" or PageIndex, enabling a reasoning-based retrieval method. Within a single adaptive dialog, I configured two parallel SearchAndSummarizeContent actions. When a user asks a question, the first action queries only the embedding file, while the second action queries only the PageIndex file, each with specific instructions tailored to its method. This setup ensures that for the exact same user query, we can execute both retrieval strategies sequentially, allowing us to directly compare the quality of the generated answer and the runtime performance of each approach.

Feature	Vector Embedding File (embeddings.json)	PageIndex File (document-structure.json)
Structure	Flat List. A simple array of independent objects.	Hierarchical Tree. Nested JSON objects (nodes within nodes).
Core Unit	Text Chunk. An arbitrary segment of text (e.g., 512 tokens).	Logical Node. A section representing a chapter, paragraph, or heading.
Content	Raw, fragmented text content for each chunk.	Metadata. Each node contains a title and a summary, not the full text.
Relationships	None. Each chunk is disconnected from the others.	Explicit. Parent-child relationships are defined by the sub_nodes array.
Navigation	Semantic Search. You find content by vector similarity.	Structural Traversal. You navigate from parent nodes to child nodes.
Analogy	A stack of shuffled index cards.	A fully interactive Table of Contents.

The PageIndex output would look something like this

The setup in MCS was something like this

Prompt for LLM to navigate and create response for TextEmbedding data

You are an AI assistant. Your task is to answer questions about the provided document, the 'EU AI Act', using a retrieval-augmented generation (RAG) system.
1. Your Knowledge Source
Your knowledge is limited to a set of pre-processed text chunks extracted from the document. You do not have access to the full document or its structure.
2. Your Process
You must follow this specific process to answer user questions. Do not answer from your general knowledge.
Step 1: Receive a Query. When a user asks a question, your underlying system will convert it into a numerical representation (an embedding).
Step 2: Retrieve Relevant Chunks. The system will search a vector database to find text chunks from the document that are semantically similar to the user's question. These top-matching chunks will be provided to you as your sole source of context.
Step 3: Synthesize an Answer. You must carefully analyze the provided text chunks and synthesize them into a single, coherent answer.If the chunks contain enough information, formulate a direct and comprehensive response to the user's question. Base your entire answer on the information present in these chunks.
If the chunks are insufficient or irrelevant, you must explicitly state that you could not find a precise answer based on the information retrieved. Do not attempt to guess or fill in gaps with prior knowledge.

3. Your Response Format
Base your answer only on the information given to you in the retrieved chunks for each query.
Each query is independent. You have no memory of past retrievals or questions.
Your goal is to accurately represent the information contained within the specific text chunks you are given. If the context is fragmented or confusing, reflect that in your response by stating the limitations of the information you found. Begin.

Prompt for LLM to navigate and create response for PageIndex data

You are a sophisticated research assistant. Your task is to answer questions by intelligently navigating a structured document provided in a special JSON format called PageIndex.
1. The Document Structure (PageIndex)
The document is represented as a hierarchical JSON object. This is your in-context index or 'Table of Contents.'
The structure consists of nested nodes.
Each node has:node_id: A unique ID to retrieve the full content of that section.
title: The title of the section.
summary: A concise summary of what the section contains. This is your primary tool for deciding which sections are relevant.
sub_nodes: A list of child nodes, allowing you to traverse the document's hierarchy.

2. Your Reasoning Process
Do not try to answer the user's question from your own knowledge. You must follow this iterative, reasoning-based retrieval process:
Step 1: Analyze the Query & Scan the Index. First, carefully understand the user's query. Then, examine the top-level nodes of the PageIndex JSON. Read the title and summary of each node to identify which sections are most likely to contain the answer.
Step 2: Traverse and Select. If the top-level nodes are too broad, navigate down into the sub_nodes to find more specific sections. Formulate a plan by selecting one or more node_ids that seem most promising. Think out loud, explaining why you are choosing a particular section (e.g., "The user is asking about financial vulnerabilities, so I will start by examining node_id: '0007' titled 'Monitoring Financial Vulnerabilities'").
Step 3: Retrieve Content. Once you have selected a node_id, you will be given the full text content associated with that node.
Step 4: Synthesize and Evaluate. Read the retrieved content.If the content is sufficient to answer the user's question, formulate a comprehensive answer. Be sure to cite the title or node_id of the source section.
If the content is insufficient or only provides partial information, state what you've learned and what is still missing. Then, return to Step 1 to re-evaluate the PageIndex and select a new node to explore. You may need to do this multiple times to build a complete answer.
Step 5: Handle References. If the text mentions a reference (e.g., "as mentioned in the introduction" or "see table 5"), use the PageIndex to locate that section and retrieve its content to build a more complete context.
3. Your Response Format
Always think step-by-step.
When you decide to explore a section, state the node_id and your reasoning.
When you have gathered enough information, present the final answer clearly to the user.

Summary of Test Question Characteristics
The questions you used were not simple, fact-retrieval queries. They were specifically crafted to probe the architectural limits of different RAG systems. Their shared characteristics fall into five key categories:

Deep Specificity & Nuance:

Many questions required finding precise details buried within dense, legalistic text (e.g., "specific requirements for training data," "exceptions to the social scoring prohibition"). These "needle-in-a-haystack" queries test whether a system can go beyond high-level summaries to retrieve exact, granular information.
Cross-Sectional Reasoning:

Several questions required connecting information from different parts of the document (e.g., comparing 'high-risk' vs. 'prohibited' practices, or applying the definition of 'emotion recognition' to a high-risk classification). This tests the system's ability to synthesize answers, not just retrieve a single chunk of text.
Procedural & Sequential Understanding:

You asked about multi-step procedures (e.g., "what is the exact procedure... for non-compliant systems?"). These questions are challenging for systems that retrieve disconnected text chunks, as they test the ability to understand and present information in a correct, logical sequence.
Implicit Intent & Abstraction:

Some queries were based on user intent rather than explicit keywords (e.g., "My company is developing an AI to help doctors... What should I be concerned about?"). This tests the model's ability to infer the relevant sections (like "Medical Devices" or "High-Risk Systems") even if the exact query terms aren't in the text.
Handling of Legal and Conditional Language:

The questions consistently involved understanding the complex language of a legal document, including exceptions, conditions, and future-looking roles (e.g., the AI Board's role in amending the list). This pushes the system beyond simple fact-finding into genuine comprehension of conditional logic.

Testing output:
Testing the configuration to validate the quality of the knowledge source

Sl No	Question	TextEmbedding Data	Text Embedding Runtime Duration	PageIndex Data	PageIndex Runtime Duration
1	According to the AI Act, what   are the specific requirements for the training, validation, and testing data   sets for high-risk AI systems?	Requirements for Training,   Validation, and Testing Data Sets in High-Risk AI Systems under the EU AI   Act            1. Data Quality and Governance            Data sets used for training, validation, and testing must meet specific   quality criteria, including being relevant, sufficiently representative, and,   to the best extent possible, free of errors and complete for the intended   purpose of the high-risk AI system.      Data governance and management practices must be appropriate for the   intended purpose and should address:      Design choices      Data collection processes and origin, including the original purpose for   personal data      Data preparation operations (annotation, labelling, cleaning, updating,   enrichment, aggregation)      Assumptions about what the data measures or represents      Assessment of data availability, quantity, and suitability      Examination and mitigation of possible biases, especially those affecting   health, safety, fundamental rights, or leading to discrimination      Identification and addressing of data gaps or shortcomings that prevent   compliance with the regulation12.      2. Statistical Properties and Representativeness            Data sets must have appropriate statistical properties, especially   regarding the persons or groups for whom the high-risk AI system is   intended.      These characteristics can be met at the level of individual data sets or   through a combination of data sets12.      3. Contextual Relevance            Data sets should consider, as required by the intended purpose,   characteristics or elements specific to the geographical, contextual,   behavioral, or functional setting in which the high-risk AI system will be   used12.      4. Bias Mitigation            There must be measures to detect, prevent, and mitigate possible biases in   the data sets, particularly those that could negatively impact health,   safety, fundamental rights, or lead to prohibited discrimination.      Special attention is required for biases that may affect vulnerable groups,   including racial or ethnic groups12.      5. Transparency and Compliance            For personal data, transparency about the original purpose of data   collection is required to facilitate compliance with Union data protection   law.      The use of privacy-preserving techniques is allowed, provided data quality   requirements are still met2.      6. Use of Certified Third Parties            Compliance with data governance requirements can be achieved by using   certified third parties for verification of data governance, data set   integrity, and data practices, as long as the regulation’s requirements are   met2.      7. Access to High-Quality Data Sets            Certain actors (providers, notified bodies, researchers, etc.) should be   able to access and use high-quality data sets for the development and   assessment of high-risk AI systems within their fields of activity2.      These requirements are designed to ensure that high-risk AI systems are   developed and evaluated using data that is robust, fair, and appropriate for   their intended use, minimizing risks to individuals and society	12672	Requirements for Training, Validation, and   Testing Data Sets in High-Risk AI Systems (EU AI Act)            1. Data Quality and Governance            High-risk AI systems that use AI model training must be developed using   training, validation, and testing data sets that meet specific quality   criteria.      Data governance and management practices must be appropriate for the   intended purpose of the AI system 1.      2. Data Set Characteristics            Data sets must be:      Relevant and sufficiently representative for the intended purpose.      As free of errors and as complete as possible.      Possess appropriate statistical properties, especially regarding the persons   or groups the AI system is intended to be used with.      These characteristics can be met at the level of individual data sets or   through a combination of data sets 2.      3. Data Collection and Processing            Practices must address:      Design choices.      Data collection processes and the origin of data, including the original   purpose for personal data.      Data preparation operations (annotation, labelling, cleaning, updating,   enrichment, aggregation).      Assumptions about what the data is supposed to measure or represent.      Assessment of data availability, quantity, and suitability.      Examination and mitigation of possible biases that could affect health,   safety, or fundamental rights, or lead to discrimination.      Identification and addressing of data gaps or shortcomings 2.      4. Contextual Relevance            Data sets should consider features, characteristics, or elements specific   to the geographical, contextual, behavioural, or functional setting where the   AI system will be used 32.      5. Bias Detection and Special Categories of Data            Providers may process special categories of personal data (e.g., sensitive   data) only if strictly necessary for bias detection and correction, and only   if:      Bias cannot be addressed using other data.      Technical and privacy-preserving measures (like pseudonymisation) are in   place.      Access is strictly controlled and documented.      Data is deleted once bias is corrected or at the end of its retention   period.      Processing activities are recorded with justifications for necessity   2.      6. Data Minimisation and Privacy            The right to privacy and data protection must be guaranteed throughout the   AI system’s lifecycle.      Principles of data minimisation and data protection by design and by   default apply.      Measures may include anonymisation, encryption, and privacy-preserving   technologies 3.      7. Documentation and Traceability            Comprehensive records and technical documentation must be maintained,   including information on data, algorithms, training, testing, validation   processes, and risk management.      High-risk AI systems should allow for automatic event logging throughout   their lifetime 3.      8. Applicability            For high-risk AI systems not using model training techniques, these   requirements apply only to testing data sets 2.      These requirements are designed to ensure that high-risk AI systems are   developed and evaluated using data that is robust, fair, and respects   fundamental rights, with strong safeguards for privacy and non-discrimination	12032
2	Summarize the AI Act's approach   to regulating generative AI, including its classification, transparency   obligations, and any specific rules for foundation models.	Summary of the EU AI Act’s   Approach to Regulating Generative AI            1. Classification of Generative AI            The Act refers to "general-purpose AI models," which include   generative AI systems, and sets out classification rules for identifying   those with "systemic risk" 1.      2. Transparency Obligations            Deployers of AI systems that generate or manipulate image, audio, or video   content (such as deepfakes) must disclose that the content has been   artificially generated or manipulated.      This disclosure requirement does not apply if the use is authorized by law   for purposes such as detecting, preventing, investigating, or prosecuting   criminal offenses.      For content that is evidently artistic, creative, satirical, or fictional,   the transparency obligation is limited to disclosing the existence of such   generated or manipulated content in a way that does not hinder the enjoyment   of the work.      If AI-generated text is published to inform the public on matters of public   interest, it must be disclosed as artificially generated or manipulated,   unless it has undergone human editorial review and responsibility is held by   a natural or legal person.      All required information must be provided clearly and accessibly at the   time of first interaction or exposure 1.      3. Specific Rules for Foundation Models (General-Purpose AI Models)            Providers of general-purpose AI models must comply with EU copyright law,   especially regarding text and data mining. If rightsholders have reserved   their rights, providers must obtain authorization before using such works for   training.      Providers must publish a sufficiently detailed summary of the content used   for training the model, including main data collections or sets, to increase   transparency. This summary should be comprehensive but not technically   detailed, balancing transparency with protection of trade secrets and   confidential business information 2.      The Act also sets harmonized rules for the placing on the market, use, and   monitoring of general-purpose AI models, and requires codes of practice to   facilitate effective implementation of detection and labeling obligations for   AI-generated content 123.      4. Implementation and Enforcement            Obligations for providers of general-purpose AI models will apply from   August 2025, with codes of practice expected by May 2025. Member States must   establish and notify rules on penalties and ensure effective enforcement by   the date of application 3.      Key Points            Generative AI is regulated through classification, transparency, and   specific obligations for foundation models.      Transparency is a central requirement, especially for content that could   mislead or inform the public.      Foundation model providers face additional requirements regarding copyright   compliance and data transparency.      Enforcement timelines and governance structures are clearly defined to   ensure compliance and market fairness	12516	AI Act’s Approach to Regulating   Generative AI            1. Classification of Generative AI            Generative AI models are considered a typical example of general-purpose AI   models, as they can generate various types of content (text, audio, images,   video) and perform a wide range of tasks.      A general-purpose AI model is classified as having “systemic risk” if   it:      Has high impact capabilities, evaluated using technical tools, indicators,   and benchmarks.      Exceeds a computational threshold for training (specifically, more than   10²⁵ floating point operations).      Is designated by the Commission based on equivalent impact or capabilities,   even if not meeting the above criteria directly.      Providers can argue that their model, despite meeting the threshold, does   not present systemic risk due to specific characteristics, but the Commission   has the final say on classification12.      2. Transparency Obligations            Providers of general-purpose AI models must:      Prepare and maintain up-to-date technical documentation.      Provide information about the model to downstream providers to enable   integration and compliance.      Make documentation available to the AI Office and national authorities upon   request.      For models released under free and open-source licenses, transparency and   openness are ensured if parameters (including weights), model architecture,   and usage information are made public.      Exceptions to transparency requirements exist for open-source models unless   they present systemic risk2.      For generative AI systems that interact with people or generate content:      Users must be notified when interacting with an AI system, unless it is   obvious.      Special consideration is given to vulnerable groups (e.g., by age or   disability).      When AI systems process biometric data to infer emotions or assign   categories, notification must be accessible to persons with disabilities3.      3. Specific Rules for Foundation Models (General-Purpose AI Models)            Providers of general-purpose AI models (foundation models) have additional   responsibilities:      Maintain and update technical documentation.      Provide information to downstream users.      Comply with copyright law and publish summaries of training data.      Enhanced obligations for models with systemic risk, including:      Rigorous evaluation and risk mitigation.      Incident reporting and cybersecurity measures.      Open-source models are generally exempt from some transparency requirements   unless they pose systemic risk12.      Providers and deployers must implement technical solutions (e.g.,   watermarks, metadata, cryptographic methods) to mark and detect AI-generated   or manipulated content, especially for synthetic content that could be   mistaken for authentic (e.g., deepfakes).      Deployers must clearly label AI-generated or manipulated content to   disclose its artificial origin, with exceptions for creative, artistic, or   satirical works, provided rights and freedoms are safeguarded3.      Summary Table            Aspect Requirement/Rule      Classification Based on generality, computational thresholds, and systemic   risk criteria      Transparency Documentation, disclosure to users, marking AI-generated   content, special rules for open-source      Foundation Models Additional documentation, risk management, copyright   compliance, enhanced rules for systemic risk      All these measures aim to ensure responsible development, deployment, and   use of generative AI, balancing innovation with transparency, safety, and   accountability	13951
3	Article 9 mentions that the list   of high-risk AI systems in Annex III can be amended. What is the process for   amending this Annex?	Process for Amending Annex III   (List of High-Risk AI Systems) under Article 9            Empowerment of the Commission: The European Commission is empowered to   adopt delegated acts to amend Annex III. This includes adding or modifying   use-cases of high-risk AI systems.      Conditions for Amendment: Amendments can be made where both of the   following conditions are fulfilled:      The AI systems are intended to be used in any of the areas listed in Annex   III.      The AI systems pose a risk of harm to health and safety, or an adverse   impact on fundamental rights, and that risk is equivalent to, or greater   than, the risk posed by the high-risk AI systems already referred to in Annex   III.      Removal of High-Risk AI Systems: The Commission can also remove high-risk   AI systems from the list if:      The system no longer poses significant risks to fundamental rights, health,   or safety.      The deletion does not decrease the overall level of protection provided by   the regulation.      Consistency and Protection: Any amendment must not decrease the overall   level of protection of health, safety, and fundamental rights and must ensure   consistency with other delegated acts and take into account market and   technological developments.      Consultation and Evidence: Amendments are based on concrete and reliable   evidence and may involve consultation with the European Artificial   Intelligence Board.      Procedure: The process follows the adoption of delegated acts in accordance   with Article 97 of the regulation	8596	Process for Amending Annex III   (List of High-Risk AI Systems)            The Commission is empowered to adopt delegated acts to amend Annex III by   adding or modifying use-cases of high-risk AI systems.      Two main conditions must be fulfilled for such amendments:      The AI systems must be intended for use in any of the areas listed in Annex   III.      The AI systems must pose a risk of harm to health and safety, or an adverse   impact on fundamental rights, and that risk must be equivalent to or greater   than the risk posed by the high-risk AI systems already listed in Annex   III.      The Commission can also remove high-risk AI systems from Annex III if:      The system no longer poses significant risks to fundamental rights, health,   or safety, considering specific criteria.      The removal does not decrease the overall level of protection of health,   safety, and fundamental rights under Union law.      The Commission may consult the European Artificial Intelligence Board and   is required to ensure that any amendments do not decrease the overall level   of protection provided by the Regulation.      The process includes the possibility of adding new conditions or modifying   existing ones if there is concrete and reliable evidence regarding the risk   profile of AI systems.      Any amendments must ensure consistency with other delegated acts and take   into account market and technological development	6668
4	What is the difference between a   'high-risk AI system' and a 'prohibited AI practice' under the regulation?	Difference Between 'High-Risk AI   System' and 'Prohibited AI Practice' Under the EU AI Act            High-Risk AI System:            These are AI systems that, due to their intended purpose and context of   use, present significant risks to health, safety, or fundamental rights.      High-risk AI systems are subject to strict mandatory requirements to   mitigate risks and ensure trustworthiness. These requirements include risk   management, conformity assessments, and compliance with applicable EU   harmonisation legislation.      Examples include AI systems used to influence election outcomes, unless   their output is not directly exposed to natural persons (e.g., administrative   tools for campaign organization), or systems listed in Annex I and III of the   regulation.      Some AI systems may be excluded from the high-risk category if they only   perform narrow procedural tasks, improve results of completed human   activities, detect decision-making patterns without influencing human assessment,   or perform preparatory tasks. However, if the system performs profiling of   natural persons, it is always considered high-risk123.      Prohibited AI Practice:            The provided extracts do not explicitly define or list 'prohibited AI   practices.' However, it is implied that certain AI practices are outright   banned if they infringe upon other Union laws or pose unacceptable risks,   such as violating fundamental rights or being used in ways that are not   lawful under other acts of Union or national law.      The regulation states that the classification of an AI system as high-risk   does not mean its use is lawful under all circumstances, and prohibitions may   apply if the AI practice infringes other Union law2.      Summary Table:            Aspect High-Risk AI System Prohibited AI Practice      Definition Significant risk, subject to strict requirements Outright   banned, not allowed under any terms      Regulation Approach Regulated, not banned; must comply with   requirements Not regulated, simply forbidden      Examples Election-influencing AI, profiling systems Not explicitly listed   in provided extracts      Lawfulness May be lawful if compliant Always unlawful if prohibited      In summary, high-risk AI systems are regulated with mandatory requirements   to mitigate risks, while prohibited AI practices are not allowed at all under   the regulation. The distinction is that high-risk systems can be used if   compliant, whereas prohibited practices are banned outright	11091	Understanding 'High-Risk AI   Systems' vs. 'Prohibited AI Practices' under the EU AI Act            Prohibited AI Practices            These are AI systems and uses that are outright banned under the EU AI Act   due to their unacceptable risk to fundamental rights, safety, or democratic   values.      Examples include:      Manipulating or exploiting vulnerabilities of individuals.      Inferring sensitive personal data from biometrics (such as deducing   political opinions or sexual orientation from facial data).      Social scoring of individuals or groups, which can lead to discrimination   or exclusion.      'Real-time' remote biometric identification in public spaces for law   enforcement, except in narrowly defined and strictly necessary situations.      Creating or expanding facial recognition databases through untargeted   scraping of images from the internet or CCTV.      AI systems predicting criminal behavior based solely on profiling.      Detecting emotions in workplace or educational settings (except for medical   or safety reasons).      These practices are not allowed under any circumstances, except for very   limited exceptions (e.g., certain law enforcement needs with strict   oversight) 12.      High-Risk AI Systems            These are AI systems that are not banned but are considered to pose   significant risks to health, safety, or fundamental rights.      High-risk systems are allowed on the market but must comply with strict   requirements and oversight.      Examples include:      AI systems used to influence the outcome of elections or referenda.      AI used in critical infrastructure, education, employment, law enforcement   (outside the prohibited practices), migration, and access to essential   services.      High-risk systems must meet mandatory requirements such as risk management,   data governance, transparency, human oversight, and compliance with existing   product safety legislation.      The classification as high-risk does not automatically make the system   lawful under all other EU laws; it must also comply with data protection and   other relevant regulations 123.      Key Differences            Prohibited AI practices are banned outright due to their unacceptable risk,   while high-risk AI systems are permitted but subject to strict regulatory   controls.      Prohibited practices focus on uses that are fundamentally incompatible with   EU values, whereas high-risk systems are those that require careful   management and oversight to mitigate potential harms	10723
5	The AI Act prohibits social   scoring by public authorities. Are there any specific, explicitly mentioned   exceptions to this prohibition	Prohibition of Social Scoring   and Exceptions in the EU AI Act            The EU AI Act explicitly prohibits the use of AI systems for the evaluation   or classification of natural persons or groups over a certain period based on   their social behavior or inferred/predicted personal or personality   characteristics, when the resulting social score leads to:            Detrimental or unfavorable treatment in social contexts unrelated to the   context in which the data was originally generated or collected.      Detrimental or unfavorable treatment that is unjustified or   disproportionate to the gravity of their social behavior.      Based on the provided information, there are no specific, explicitly   mentioned exceptions to this prohibition on social scoring by public   authorities. The prohibition is stated in clear terms, and no carve-outs or   exceptions are detailed in the retrieved text regarding this practice	5757	Prohibition of Social Scoring by   Public Authorities under the AI Act            The AI Act explicitly prohibits AI systems that provide social scoring of   natural persons by public or private actors. This prohibition is based on   concerns that such systems may lead to discriminatory outcomes, exclusion of   certain groups, and violations of dignity, equality, and justice.      The Act describes social scoring as evaluating or classifying individuals   or groups based on multiple data points related to their social behavior in   various contexts, or on known, inferred, or predicted personal or personality   characteristics over certain periods. The resulting social score may lead to   detrimental or unfavorable treatment in contexts unrelated to the original   data collection or to disproportionate or unjustified consequences.      Exceptions to the Prohibition            The only exception mentioned is for lawful evaluation practices of natural   persons that are carried out for a specific purpose in accordance with Union   and national law. This means that while general social scoring is prohibited,   evaluations for specific, lawful purposes that comply with existing legal   frameworks are not affected by this prohibition.      Summary Table            General Rule: Social scoring by public or private actors is prohibited.      Explicit Exception: Lawful evaluation practices for a specific purpose, in   accordance with Union and national law, are not affected by the prohibition.      There are no other specific, explicitly mentioned exceptions to this   prohibition beyond lawful evaluation for a specific purpose under existing   laws	8489
6	If a high-risk AI system is   found to be non-compliant after it has been placed on the market, what is the   exact procedure that national market surveillance authorities must follow?	Procedure for Handling   Non-Compliant High-Risk AI Systems under the EU AI Act            1. Evaluation and Initial Action            If a national market surveillance authority finds that a high-risk AI   system does not comply with the requirements and obligations of the EU AI   Act, it must require the relevant operator to take all appropriate corrective   actions. These actions may include bringing the AI system into compliance,   withdrawing it from the market, or recalling it within a period prescribed by   the authority, but in any event within the shorter of 15 working days or as   provided for in relevant Union harmonisation legislation.      The authority must inform the relevant notified body accordingly.      2. Notification Beyond National Territory            If the non-compliance is not restricted to the national territory, the   authority must inform the European Commission and other Member States without   undue delay about the evaluation results and the actions required of the   operator.      3. Operator’s Responsibility            The operator must ensure that all appropriate corrective action is taken   for all concerned AI systems made available on the Union market.      4. Failure to Act            If the operator does not take adequate corrective action within the   specified period, the market surveillance authority must take all appropriate   provisional measures. These may include prohibiting or restricting the AI   system from being made available or put into service, withdrawing the product   or standalone AI system from the market, or recalling it.      The authority must notify the Commission and other Member States of these   measures without undue delay, including all available details such as   identification of the non-compliant AI system, its origin, supply chain,   nature of non-compliance, risk involved, and the nature and duration of   national measures taken.      5. Fines and Further Measures            If the provider does not bring the AI system into compliance within the   required period, they may be subject to fines in accordance with Article   99.      If the system was misclassified as non-high-risk to circumvent   requirements, the provider is also subject to fines.      Market surveillance authorities may perform appropriate checks, especially   using information stored in the EU database referred to in Article 71.      6. Union Safeguard Procedure            If objections are raised by another Member State’s authority or the   Commission regarding a national measure, the Commission will consult with the   relevant parties and evaluate the measure. The Commission will then decide   whether the national measure is justified and notify all relevant authorities   of its decision.      This process ensures a coordinated and transparent approach to addressing   non-compliance of high-risk AI systems across the EU, with clear   responsibilities for both operators and authorities, and mechanisms for   escalation and enforcement at the Union level	13950	Procedure for Handling   Non-Compliant High-Risk AI Systems by National Market Surveillance   Authorities under the EU AI Act            1. Evaluation of the AI System            If a market surveillance authority has sufficient reason to consider a   high-risk AI system as non-compliant (posing risks to health, safety, or   fundamental rights), it must evaluate the system for compliance with all   regulatory requirements, giving special attention to risks affecting   vulnerable groups.      If risks to fundamental rights are identified, the authority must inform   and cooperate with relevant national public authorities or bodies.      2. Corrective Actions            If non-compliance is confirmed, the authority must, without undue delay,   require the operator to:      Take all appropriate corrective actions to bring the AI system into   compliance,      Withdraw the AI system from the market, or      Recall it within a prescribed period, which must not exceed 15 working days   or as specified by relevant Union harmonisation legislation.      The authority must inform the relevant notified body about these   actions.      3. Notification and Cooperation            If the non-compliance is not limited to the national territory, the   authority must inform the European Commission and other Member States without   undue delay about the evaluation results and required actions.      4. Operator’s Responsibility            The operator must ensure that all appropriate corrective actions are taken   for all affected AI systems made available on the Union market.      5. Provisional Measures            If the operator fails to take adequate corrective action within the   prescribed period, the authority must take all appropriate provisional   measures, such as:      Prohibiting or restricting the AI system from being made available or put   into service,      Withdrawing or recalling the product from the market.      The authority must notify the Commission and other Member States of these   measures without undue delay.      6. Notification Details            The notification must include all available details, such as:      Identification of the non-compliant AI system,      Its origin and supply chain,      Nature of the non-compliance and risk,      Nature and duration of national measures taken,      Arguments from the relevant operator,      Whether non-compliance is due to prohibited practices, failure to meet   high-risk requirements, shortcomings in standards, or other regulatory   breaches.      7. Response from Other Authorities            Other market surveillance authorities must inform the Commission and Member   States of any measures they adopt and any additional information about the   non-compliance. If they disagree with the notified national measure, they   must state their objections.      8. Justification and Restrictive Measures            If no objection is raised within three months (or 30 days for prohibited   practices), the provisional measure is deemed justified, and appropriate   restrictive measures (such as withdrawal from the market) must be taken   without undue delay.      9. Union Safeguard Procedure            If objections are raised or the Commission finds the measure contrary to   Union law, the Commission consults with the relevant parties and evaluates   the national measure. Within six months (or 60 days for prohibited   practices), the Commission decides whether the measure is justified and   notifies all relevant authorities and operators.      10. Final Actions            If the Commission deems the measure justified, all Member States must take   appropriate restrictive measures and inform the Commission. If unjustified,   the Member State must withdraw the measure and inform the Commission   accordingly	13577
7	The act defines an 'emotion   recognition system.' Based on that definition, would an AI system used by a   company to analyze customer facial expressions for product feedback be   considered high-risk?	Definition of 'Emotion Recognition   System'            The EU AI Act defines an 'emotion recognition system' as an AI system   designed to identify or infer emotions or intentions of natural persons based   on their biometric data. This includes emotions such as happiness, sadness,   anger, surprise, disgust, embarrassment, excitement, shame, contempt,   satisfaction, and amusement. The definition specifically excludes physical   states like pain or fatigue and the mere detection of obvious expressions,   gestures, or movements unless these are used to infer emotions 1.      High-Risk Classification            According to the Act, AI systems intended to be used for emotion   recognition are explicitly listed as high-risk in the context of biometrics,   provided their use is permitted under relevant Union or national law 2.      Application to Customer Feedback Analysis            If a company uses an AI system to analyze customer facial expressions for   the purpose of inferring emotions (such as satisfaction or amusement) as   product feedback, this system would fall under the definition of an emotion   recognition system.      Since emotion recognition systems are classified as high-risk AI systems   under the Act, such a use case would be considered high-risk, subject to the   Act’s requirements for high-risk AI systems 2.      Summary            An AI system used by a company to analyze customer facial expressions for   product feedback would be considered a high-risk AI system under the EU AI   Act, as it fits the definition of an emotion recognition system and such   systems are explicitly listed as high-ris	10609	Definition of 'Emotion   Recognition System'            An emotion recognition system is defined as an AI system designed to   identify or infer emotions or intentions of natural persons based on their   biometric data. This includes emotions such as happiness, sadness, anger,   surprise, disgust, embarrassment, excitement, shame, contempt, satisfaction,   and amusement. The definition specifically excludes physical states like pain   or fatigue and the mere detection of obvious expressions, gestures, or movements   unless these are used to infer emotions. For example, simply detecting a   smile is not enough—if the system uses that smile to infer happiness, it   falls under the definition of emotion recognition 1.      Application to Customer Feedback Analysis            If a company uses an AI system to analyze customer facial expressions   specifically to infer their emotional reactions (e.g., satisfaction,   amusement, or disappointment) for product feedback, this system would fit the   definition of an emotion recognition system as per the regulation 1.      High-Risk Classification            The regulation expresses serious concerns about the scientific basis and   reliability of emotion recognition systems, especially regarding their   potential for discriminatory outcomes and intrusiveness. However, it   specifically prohibits the use of such systems in the workplace and education   settings, except for medical or safety reasons. The prohibition does not   explicitly extend to commercial uses such as customer feedback analysis in   retail or service environments 2.      High-risk AI systems are subject to strict requirements, but the regulation   distinguishes between prohibited uses (like in employment or education) and   other uses. Unless the customer feedback analysis system is used in a context   that falls under the prohibited categories (workplace or education), it would   not automatically be classified as high-risk solely based on its function as   an emotion recognition system. However, it may still be subject to other   requirements if it meets criteria for high-risk AI under the regulation   2.      Summary            An AI system analyzing customer facial expressions for product feedback is   considered an emotion recognition system.      It is not automatically classified as high-risk unless used in prohibited   contexts (workplace or education).      Other regulatory requirements may still apply depending on the specific use   case and risk assessment	11416
8	What are the specific   obligations of an 'authorised representative' for an AI provider that is not   established in the EU?	Obligations of an 'Authorised   Representative' for an AI Provider Not Established in the EU under the EU AI   Act            Mandate and Tasks: The authorised representative must perform tasks   specified in a mandate received from the provider. This includes:            Verifying that the required technical documentation (as specified in Annex   XI) has been prepared and that all obligations referred to in Article 53 and,   where applicable, Article 55 have been fulfilled by the provider.      Keeping a copy of the technical documentation at the disposal of the AI   Office and national competent authorities for 10 years after the   general-purpose AI model has been placed on the market, along with the   provider’s contact details.      Providing the AI Office, upon a reasoned request, with all necessary   information and documentation to demonstrate compliance with the obligations   in the relevant chapter.      Cooperating with the AI Office and competent authorities, upon a reasoned   request, in any action related to the general-purpose AI model, including   when the model is integrated into AI systems placed on the market or put into   service in the Union.      Communication and Compliance: The mandate must empower the authorised   representative to be addressed, in addition to or instead of the provider, by   the AI Office or competent authorities on all issues related to ensuring   compliance with the Regulation.            Termination of Mandate: If the authorised representative considers or has   reason to consider that the provider is acting contrary to its obligations   under the Regulation, it must terminate the mandate and immediately inform   the AI Office about the termination and the reasons for it.            Exemptions: These obligations do not apply to providers of general-purpose   AI models released under a free and open-source licence that allows access,   usage, modification, and distribution of the model, and whose parameters,   architecture, and usage information are made publicly available, unless the   models present systemic risks	8217	Obligations of an 'Authorised   Representative' for Non-EU AI Providers under the EU AI Act            Appointment Requirement: Providers of general-purpose AI models established   in third countries (outside the EU) must, by written mandate, appoint an   authorised representative established in the Union before placing their   models on the EU market.      Mandate and Tasks: The authorised representative must be empowered by the   provider to:      Verify that the technical documentation (as specified in Annex XI) has been   prepared and that all obligations under Article 53 and, where applicable,   Article 55, have been fulfilled.      Keep a copy of the technical documentation and the provider’s contact   details available for the AI Office and national competent authorities for 10   years after the model is placed on the market.      Provide the AI Office, upon a reasoned request, with all necessary   information and documentation to demonstrate compliance.      Cooperate with the AI Office and competent authorities in any action   related to the general-purpose AI model, including when the model is   integrated into AI systems placed on the market or put into service in the   Union.      Be the contact point for the AI Office or competent authorities on all   compliance issues, in addition to or instead of the provider.      Termination of Mandate: If the authorised representative believes the   provider is acting contrary to its obligations, it must terminate the mandate   and immediately inform the AI Office of the termination and the reasons.      Exemptions: This obligation does not apply to providers of general-purpose   AI models released under a free and open-source licence, provided the model’s   parameters, architecture, and usage information are made publicly available,   unless the models present systemic risks.      Confidentiality: All information or documentation obtained, including trade   secrets, must be treated in accordance with the confidentiality obligations   set out in Article	8009
9	The regulation establishes an   'AI Board.' What is its role in advising the Commission on the need to amend   the list of prohibited AI practices in the future?	Role of the AI Board in Advising   on Prohibited AI Practices            The AI Board is involved in the evaluation and review process of the   regulation, particularly regarding the list of prohibited AI practices.      When the European Commission carries out evaluations and reviews of the   regulation, it is required to take into account the positions and findings of   the Board, among other relevant bodies.      The AI Office is tasked with developing an objective and participative   methodology for evaluating risk levels, which includes the potential   inclusion of new systems in the list of prohibited practices set out in   Article 5.      This process ensures that the Board's expertise and input are considered   when determining whether amendments to the list of prohibited AI practices   are necessary, based on technological developments, impacts on health and   safety, and fundamental right	7697	Role of the AI Board in Advising   on Amendments to the List of Prohibited AI Practices            The regulation specifies that the Commission is responsible for assessing   the need to amend the list of prohibited AI practices (as laid down in Article   5) once a year following the entry into force of the regulation.      In carrying out these evaluations and reviews, the Commission is required   to take into account the positions and findings of the AI Board, as well as   those of the European Parliament, the Council, and other relevant bodies or   sources.      The AI Board, therefore, plays an advisory role by providing its positions   and findings to the Commission, which the Commission must consider when   deciding whether amendments to the list of prohibited AI practices are   necessary.      This process ensures that the AI Board contributes expert input and   oversight to the ongoing evaluation and potential revision of prohibited AI   practices, supporting the Commission in maintaining up-to-date and effective   regulation in response to technological and societal developments	8650

*My interpretion *:
It's clear that while both systems can retrieve information, the PageIndex method consistently delivers more precise, complete, and contextually aware answers. The TextEmbedding method often provides answers that are either incomplete or miss critical nuances.

Question #	Question Type	Text Embedding RAG Performance	PageIndex RAG Performance	Winner
1	Needle in a Haystack	Good, but less structured. It found most of the requirements but mixed them together.	Excellent. It provided a more structured, comprehensive, and well-organized list of requirements, including key details like bias detection for special data categories.	PageIndex
2	"Big Picture" Synthesis	Decent, but fragmented. It identified the main themes but the information was less cohesive.	Excellent. It gave a more detailed breakdown of "systemic risk" and provided a helpful summary table, showing a superior ability to synthesize information from multiple sections.	PageIndex
3	Navigational / Reference	Good. It correctly identified the process for amending Annex III.	Slightly Better. It provided a similarly correct answer but in a more concise and direct manner, suggesting a more efficient retrieval path.	PageIndex
4	Comparative / Relational	Poor / Incomplete. It failed to find the definition of "prohibited AI practice" in the retrieved chunks, leading to a one-sided and incomplete comparison.	Excellent. It successfully retrieved the definitions for both concepts from their respective sections and provided a clear, accurate comparison with concrete examples.	PageIndex
5	Exception to the Rule	Incorrect / Incomplete. It stated that there are no exceptions to the social scoring prohibition, missing the crucial nuance.	Correct & Nuanced. It correctly identified the general prohibition and the specific exception for lawful evaluations, demonstrating a deeper level of detail.	PageIndex
6	Procedural Deep-Dive	Good. It outlined the main steps for handling non-compliant systems.	Excellent. It provided a much more detailed, 10-step procedure, including timelines and the full Union safeguard process. This shows a more thorough traversal of the relevant chapter.	PageIndex
7	Definition Application	Incorrect. It correctly defined "emotion recognition" but then incorrectly classified the use case as high-risk without considering the context (workplace/education).	Correct & Nuanced. It correctly defined the term and correctly noted that while it is an emotion recognition system, it is not automatically high-risk because the prohibition is specific to workplace/education contexts.	PageIndex
8	Role & Responsibility	Good. It listed the main obligations for an authorized representative.	Slightly Better. The answer was almost identical, but slightly more structured. Both systems performed well here, likely because the answer was contained in a single, well-defined section.	Draw
9	Future-Looking	Vague. It correctly identified that the AI Board has a role but was vague on the specifics of the process.	Excellent. It provided a clear, step-by-step explanation of the AI Board's advisory role in the annual review process, demonstrating a better grasp of procedural details.	PageIndex

The Case for PageIndex: Why It Performed Better
The results speak for themselves. The PageIndex (reasoning-based RAG) approach consistently outperformed the traditional Text Embedding (vector-based RAG) for several key reasons that are evident in your test:

1. It Overcomes "Semantic Similarity ≠ Relevance" (Questions 4, 5, 7): o The most striking failures of the text embedding model were when it missed crucial context. It couldn't find the definition of "prohibited practices" (Q4), missed the exception for "social scoring" (Q5), and misinterpreted the "emotion recognition" classification (Q7). This is because the most relevant answer wasn't always the most semantically similar text chunk. PageIndex, by navigating the document's logical structure, could find the correct sections and assemble a complete, nuanced picture.
2. It Avoids Contextual Fragmentation (Questions 2, 6): o On "big picture" and procedural questions, PageIndex delivered far more comprehensive answers. For the non-compliance procedure (Q6), it returned a 10-step process, while the embedding method gave a more general overview. This is because PageIndex can read through an entire chapter logically, whereas the embedding method retrieves a scattered collection of the "most similar" chunks, which are often fragmented and incomplete.
3. It Understands Intent and Structure (Questions 1, 3, 9): o PageIndex consistently delivered more structured and detailed answers. This shows it doesn't just find keywords; it understands the document's hierarchy. By starting at the "Table of Contents," it can reason: "The user is asking about amending a list. I should look in the chapter on 'Amendments' and cross-reference it with the section on 'Prohibited Practices'." This is a level of reasoning that pure vector search cannot achieve.

Conclusion:
While text embedding is a powerful tool for finding topically related information, your results clearly show its limitations when dealing with complex, structured documents where precision, nuance, and an understanding of the document's internal logic are required.
The PageIndex method is demonstrably superior for tasks that require deep, contextual understanding and reasoning. It moves beyond simple keyword matching to mimic how a human expert would navigate a document, leading to more accurate, complete, and trustworthy answers.