This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
What I Built
Your business changed. Did every document get the memo?
Imagine you're a freelancer running a small web design agency.
You update the price of your starter package from ₹15,000 to ₹18,000. You change your pricing page, update your spreadsheet, and move on.
Except your old price is still sitting in a PDF proposal template. Your FAQ still mentions the old package. An outdated quotation is waiting in your drafts.
A few weeks later, you discover the problem when a client points it out.
The frustrating part? Your information wasn't missing. It was simply inconsistent.
That's the problem I'm tackling with CAUSEWEFT: an open-source AI tool that helps small businesses discover information drift across documents and understand what needs to change when an important business fact changes.
The idea is simple:
Change one fact. Find its ripple effects. Fix them with evidence.
How CAUSEWEFT works
CAUSEWEFT is designed around four steps:
1. Gather the scattered information
Import business documents such as pricing sheets, proposals, product catalogs, FAQs, policies, and website content.
2. Extract facts, not just summaries
Instead of only summarizing a document, the AI extracts structured claims: what a product costs, what a service includes, when a deadline applies, and which source supports each claim.
3. Find information drift
CAUSEWEFT compares related claims across different sources. It can flag conflicting prices, inconsistent service descriptions, outdated dates, and other discrepancies.
Each finding should point back to the evidence behind it. An uncertain relationship is marked for review instead of being presented as a confirmed contradiction.
4. Propose a coordinated repair
When a fact changes, CAUSEWEFT identifies potentially affected documents and prepares suggested edits.
The user reviews and approves the changes. The system then checks whether the inconsistencies have actually been resolved.
The goal isn't to let AI rewrite a business's records without supervision. It's to make hidden inconsistencies visible and the work of fixing them much less tedious.
Who would pay for this?
I'm starting with freelancers, small agencies, independent consultants, and online sellers.
These businesses often maintain important information across documents and applications without having a dedicated operations team. A single outdated price, incorrect policy, or conflicting promise can mean lost revenue, avoidable disputes, or wasted time.
CAUSEWEFT could eventually become a paid tool for keeping that information consistent, with a free open-source core and optional paid integrations or team features.
For the first version, I'm focusing on one narrow workflow: upload a small collection of business documents, identify conflicting facts, inspect the source evidence, and generate a reviewable repair plan.
Demo
Try the working prototype: [INSERT DEPLOYED DEMO URL]
Watch the walkthrough: [INSERT SHORT DEMO VIDEO URL]
The demo should use a sample business with a changed service price and several deliberately outdated references. The viewer should be able to follow the chain from the original fact to the detected inconsistencies and the proposed repairs.
Code
GitHub repository: [INSERT PUBLIC GITHUB REPOSITORY URL]
The repository will include the application code, setup instructions, sample data, and instructions for running the model locally.
Contributions that improve document parsers, claim extraction, contradiction detection, and validation rules will be especially valuable.
How I Built It
CAUSEWEFT is designed as a combination of open-weight AI and deterministic validation rather than a chatbot wrapped around document uploads.
The planned architecture has four main components:
- Document ingestion: Python-based parsers extract text and relevant metadata from supported files.
- AI-powered claim extraction: Google's open-weight Gemma model, served locally through Ollama, converts unstructured content into structured claims and identifies potentially related facts.
- Evidence and consistency engine: A structured local store preserves extracted claims, their source locations, and relationships. Deterministic checks handle comparisons such as prices and dates, while ambiguous semantic conflicts are flagged for review.
- Review and repair workflow: A lightweight web interface presents each discrepancy, the supporting evidence, and a proposed correction. Changes require approval before they are applied.
The key architectural decision is to keep fact extraction separate from verification.
An LLM can suggest that two statements conflict, but the application should preserve the original evidence, validate what it can deterministically, and distinguish confirmed discrepancies from uncertain interpretations.
The first implementation will focus on supported document formats and a small set of testable inconsistency types rather than pretending to understand every kind of business information.
[UPDATE THIS SECTION TO MATCH THE ACTUAL IMPLEMENTATION BEFORE PUBLISHING. Retain Gemma and Ollama only if they are genuinely integrated into the working project.]
Why Does Open Innovation Matter?
Business documents often contain sensitive information: customer details, quotations, contracts, pricing, and internal policies.
I want CAUSEWEFT to make local AI a practical option rather than forcing users to send every document to a proprietary API.
Using an open-weight model gives the project room to run inference locally, inspect and modify the extraction pipeline, experiment with different models, and adapt prompts and validation strategies for different industries.
It also makes the project more extensible. Someone building an invoicing application could contribute a connector. Someone working with product catalogs could add specialized consistency rules. A freelancer could adapt the workflow to their own documents.
The model is only one part of the system. The open-source value comes from making the full process—claim extraction, evidence tracking, validation, and repair—available for people to inspect, test, and improve.
My goal is to make CAUSEWEFT useful even when users don't trust an AI model to make decisions on their behalf.
AI should help discover the problem. Evidence should explain it. The user should remain in control of the fix.
Prize Categories
- Best Use of Gemma — [KEEP ONLY IF GEMMA IS MEANINGFULLY USED IN THE IMPLEMENTED PROJECT.]
What's Next?
I want to expand CAUSEWEFT beyond detecting outdated prices.
The longer-term direction is a reusable consistency engine that can reason about relationships across different information sources, identify the consequences of a change, and help users coordinate safe updates.
For now, the priority is much more concrete: prove that CAUSEWEFT can detect real inconsistencies in a realistic set of business documents, show exactly why each finding was raised, and help the user fix them without introducing new errors.
If you've ever changed something in one place only to discover an outdated version somewhere else, that's the problem I'm building this for.
I'd especially love feedback on which kinds of information drift cost small businesses the most time or money.
Top comments (0)