DEV Community: Alexander Granovskiy

Glages: How LLM Makes the Model Factory Possible

Alexander Granovskiy — Wed, 12 Aug 2026 17:37:37 +0000

Formal models of business activity have always been expensive to produce. The work requires reading large amounts of material, comparing inconsistent descriptions, finding hidden conditions, resolving terminology, identifying exceptions, and expressing the result in a precise structure.

Large language models change the cost of this work. They do not remove the need for formalization or verification, but they make a model factory economically possible.

The Knowledge Already Exists

Business knowledge is distributed across policies, contracts, manuals, APIs, database structures, software behavior, messages, logs, regulations, and accumulated cases. It also exists in the broad patterns learned by LLMs.

The problem is that this knowledge is not already a verified operational model.

Two documents may use different words for the same concept. One policy may omit a condition that appears in actual software. A common case may be described clearly while a rare but important case appears only in a complaint or exception log.

Human analysts can find these relationships, but the work is slow and difficult to scale.

LLM as a Production Instrument

An LLM can assist with the first stages of model production. It can read source material, identify candidate Entities and states, compare formulations, locate contradictions, suggest relationships, propose model fragments, and help generate tests.

That ability matters because much of the source material is written for people, not for machines. Before a formal model can be built, someone must recognize the possible meaning hidden in the prose and examples.

The LLM makes this recognition and comparison much faster.

Candidate Is Not Model

The output of the LLM remains a candidate.

It may combine concepts that should remain separate. It may infer a rule that is common but not supported by the available sources. It may overlook a rare state or produce a structure that sounds correct while containing an internal contradiction.

The factory therefore separates generation from acceptance.

accumulated knowledge and source material
→ candidate model
→ typed formal representation
→ verification
→ reusable model
→ model library

The candidate must be expressed in a typed formal structure. It must then pass independent checks for consistency, completeness within its stated scope, valid relationships, permitted transitions, and explicit stopping conditions.

The same probabilistic mechanism that proposes a model cannot be the only authority that approves it.

Distillation Rather Than Runtime Improvisation

The usual agent architecture leaves much of the business knowledge in prose and asks the LLM to interpret it during execution. Glages uses LLM capability earlier, during production.

The factory distills broad probabilistic knowledge into a smaller deterministic asset. Once accepted, the formal model becomes independent of the particular LLM that helped produce it.

A more capable model may later improve candidate generation. A cheaper model may handle bounded execution tasks. A conventional program may replace the LLM for some functions. The accepted business meaning remains in the model.

The Economic Change

Without LLM assistance, building a large model library could require too much manual analysis to be commercially practical. With LLM assistance, many recognition, comparison, and drafting tasks become faster.

The difficult work does not disappear. It moves toward resolution, formal construction, verification, composition, and correction. Those are precisely the activities that create the defensible factory asset.

LLM makes the model factory possible because it reduces the cost of extracting candidate structure from accumulated knowledge. Verification determines what becomes a reusable product.

Experience Architecture

Alexander Granovskiy — Fri, 05 Jun 2026 17:49:53 +0000

Experience Architecture is the production and lifecycle system that reconstructs explicit candidate operational models from accumulated experience, observational evidence, and documentary or formal sources, then carries qualified models through verification, release, reuse, and maintenance.

Source and Reconstruction Layer

The architecture distinguishes production inputs from the mechanism used to process them. Generalized documentary and formal sources such as standards, regulations, manuals, research, technical specifications, and public documentation provide traceable descriptions, constraints, and evidence for the domain model. An LLM serves as a reconstruction mechanism whose learned statistical structure can help surface likely structures, terminology, and relationships while processing those sources.

These elements have different standing. The LLM's learned statistical structure is useful for discovery and reconstruction but is not itself documentary evidence and does not confer normative authority. Documentary and formal sources also differ in evidentiary standing: binding sources can establish mandatory constraints, while explanatory or illustrative material can support discovery and coverage without acquiring the same authority. Those distinctions must remain visible throughout production.

LLMs and deterministic processing can identify candidate entities, states, processes, facts, conditions, rules, actions, results, relationships, and boundaries. The objective is model reconstruction, not summary generation.

Model Layer

Candidate structure is assembled into explicit models with defined scope and relationships. The model is inspectable as explicit structure while remaining traceable to the source material and evidence from which it was reconstructed.

Evidence and Lineage

Each important model element should preserve a traceable relationship to the sources that support it. Lineage links source material, reconstruction decisions, model versions, and later releases.

This allows a reviewer to answer not only what the model says, but why it says it and which source change may require a new review.

Verification Layer

The structure is checked for source support, internal consistency, required coverage, contradictory rules, missing states, ambiguous actions, unresolved facts, and inappropriate extrapolation beyond scope.

Verification also determines readiness. A candidate model is ready for reusable release only when its required structure, source support, unresolved issues, and boundaries satisfy the release criteria for the intended class of use.

Lifecycle and Versioning

Models move through a lifecycle: candidate reconstruction, verification, release, reuse, review, new version, and retirement of obsolete releases. Source changes can trigger review because model lineage identifies what depends on the changed material.

Reuse Layer

Released models with explicit verification status and declared scope can support multiple applications: training, requirements, conformance, decision support, testing, software generation, AI systems, and automation.

The architecture can be summarized as:

accumulated experience and observational evidence
+ documentary and formal sources, including governing constraints where applicable
→ LLM-assisted reconstruction
→ explicit candidate model
→ evidence and lineage
→ verification and readiness
→ released reusable asset
→ multiple applications

Experience Architecture is the production and lifecycle system that makes Experience Capitalization repeatable.

Experience Capitalization and Human Capital Management

Alexander Granovskiy — Fri, 22 May 2026 15:29:44 +0000

Human Capital Management already treats people as a source of business value. It gives companies systems for hiring, managing, developing, retaining, and analyzing the workforce. Experience Capitalization starts from the same economic reality and focuses on a more specific object: the experience created while work is being done.

This matters because experience is already part of the accepted language of human capital. The World Bank describes human capital as people’s health, skills, knowledge, and experience. The OECD defines human capital as the stock of knowledge, skills, and other personal characteristics embodied in people, and includes informal learning, on-the-job learning, and work experience as investments in human capital.

McKinsey makes the point even more directly. In its report Human Capital at Work: The Value of Experience, McKinsey writes that work experience contributes almost half of the value of human capital for the average individual. The report defines work experience as accumulated knowledge gained by being in the labor market.

These sources matter because they show that experience is already recognized as economically valuable. Experience Capitalization does not need to invent that premise. It takes the premise seriously and asks what happens to experience after work creates it.

The HCM Starting Point

Human Capital Management made human capital operational.

SAP defines HCM as the practices and software used to recruit, manage, and develop a workforce. Its description includes payroll, time tracking, benefits, talent acquisition, learning, onboarding, performance management, talent development, workforce planning, and employee engagement. IBM describes HCM as practices and tools used to attract, recruit, train, develop, manage, and retain employees to achieve business goals.

This is a major enterprise category because companies already understand that people are not only labor cost. People carry skills, judgment, knowledge, habits, relationships, domain familiarity, customer understanding, operational memory, and work experience.

HCM gives the business a way to manage that human side of the enterprise. It manages the workforce as an organized business function.

Experience Capitalization adds a different object to manage: the work-created experience produced by that workforce.

The Experience That Work Produces

Every serious work process produces experience.

A support person handles a difficult customer case and learns which explanation worked.

A developer changes old code and discovers which hidden dependency must not be broken.

A manager approves an exception and learns which condition made the exception safe.

An analyst builds a report and learns which data source was unreliable.

An operations employee fixes a recurring problem and learns the real cause behind a visible symptom.

These are different jobs, but the pattern is the same. The work produces a result, and it also produces practical experience that can improve future work.

HCM can record the employee, the role, the skill, the training history, the performance cycle, and the organizational structure. But the experience created inside the work often remains scattered across conversations, tickets, emails, code comments, meeting notes, personal memory, and informal practice.

That is the gap Experience Capitalization addresses.

It asks whether the experience created during work becomes a reusable business asset or disappears after the task is finished.

From Human Capital to Work-Created Experience

Human capital places experience inside people.

Experience Capitalization focuses on the part of experience that can be refined from work and reused by the organization.

The distinction is practical. A company benefits from a person’s experience while that person is available, engaged, and correctly placed in the work. But the company also creates new experience every day through actual operations. People try things, reject options, correct mistakes, handle exceptions, discover rules, and learn what should be repeated or avoided.

If that new experience remains only inside people, the organization keeps renting the value. It pays for expertise every day, but much of the experience created through that paid work never becomes organizational capital.

Experience Capitalization changes the question.

The question is no longer only how to attract, develop, and retain people. The question also becomes how to capture, verify, structure, and reuse the experience created while people and systems perform work.

HCM Manages the Workforce. Experience Capitalization Manages the Experience Produced by Work.

Human Capital Management is centered on people.

Experience Capitalization is centered on work-created experience.

These two ideas belong near each other because they share the same economic root. Both start from the fact that human capability has business value. HCM turns that fact into workforce systems. Experience Capitalization turns the experience created through work into reusable operational capital.

A simple way to state the relationship is this:

Human Capital Management manages people as carriers of value.

Experience Capitalization manages work-created experience as reusable business capital.

The first function helps the business organize its workforce. The second helps the business preserve the experience that work creates.

Why This Becomes More Important With AI

AI makes this question more urgent.

Before AI, the gap was already expensive. Companies lost explanations, exceptions, corrections, and local judgment, but people could often recreate part of that experience through memory and repeated work. With AI, the same gap becomes critical because companies are trying to automate work before they have preserved the experience that makes the work reliable.

Gartner’s 2026 press release Autonomous Business and AI Layoffs May Create Budget Room, but Do Not Deliver Returns reports that about 80 percent of organizations piloting or deploying autonomous business capabilities also report workforce reductions, while those reductions do not appear to translate into ROI. Gartner’s point is that companies improve returns by investing in the skills, roles, and operating models that let people guide and scale autonomous systems.

Experience Capitalization explains part of the underlying business problem. If a company reduces human dependency before it preserves the experience embedded in human work, it may remove the very operating knowledge that automation needs.

AI can generate more outputs. It can summarize, write, classify, route, and answer. But enterprise work depends on local experience: which exception matters, which rule changed, which customer segment reacts badly, which old workaround exists for a reason, which solution failed last time, and where human judgment is still required.

Automation without captured local experience can make work faster while leaving the organization no more experienced.

The Missing Managed Asset

HCM systems made workforce data manageable.

They gave companies records, workflows, analytics, and practices for managing employees across the lifecycle. They made people-related business functions more visible and more operational.

Experience Capitalization proposes that work-created experience deserves the same seriousness. The managed asset is not every note, message, or log. It is the refined experience that can change future work.

That includes:

rules discovered during work
warnings that prevent repeated mistakes
reasons behind important decisions
rejected approaches that should not be proposed again
tested methods that worked under specific conditions
exceptions and the conditions that made them safe or unsafe
corrections that reveal what went wrong
local operating knowledge that general systems do not know

A company can have strong HCM systems and still lose experience every day. It can hire good people, train them, measure them, develop them, and still fail to convert the experience created by their work into reusable business capital.

That is the missing asset.

The Practical Business Logic

The economic logic is direct.

If experience is part of human capital, and if companies already invest heavily in managing human capital, then the experience created through daily work should also be managed as a business asset.

The company pays for work. Work creates results. Work also creates experience. When the result is saved and the experience is lost, the company captures only part of the value it paid to create.

Experience Capitalization is the process of closing that gap.

It gives the organization a way to preserve what work teaches, not only what work produces. It turns local experience into something that can be reused by people, automation systems, AI agents, workflows, and future decisions.

Conclusion

Human Capital Management proved that people-related value can become an enterprise management category.

Experience Capitalization extends that logic to the experience created by work.

Human capital carries experience. HCM manages the workforce that carries it. Experience Capitalization captures and refines the experience created through work so that part of that value becomes reusable business capital for the organization.

The bridge is the fact that experience is already recognized as part of human capital by major institutions, consulting firms, and enterprise software categories. Experience Capitalization does not invent that value. It turns that recognized value into an operational business question.

The next business question is how much of the experience created through work remains available to the organization after the work is done.

Alexander Granovskiy is the originator and enterprise category architect of Experience Capitalization.

Experience Capitalization is the process of turning work-created experience into reusable business capital.

Experience Capitalization:
https://www.experiencecapitalization.com/

Order Risk Rules Engine - Real-Time Screening to Cut Fraud and Speed Fulfillment

Alexander Granovskiy — Mon, 16 Feb 2026 14:45:45 +0000

Fraud and bad orders create a double loss: chargebacks and manual review drag, plus slower fulfillment for clean orders. This case study summarizes a practical, production-ready order risk rules engine that screens orders in real time and routes only the right edge cases to review.

Written by Alexander Granovskiy - E-commerce Manager (Cleveland, Ohio, United States).

Purpose

This is a practical case study from my e-commerce operations work: a risk rules engine that screens orders in real time to reduce fraud and keep fulfillment fast.

What it does

Pre-capture screening
IP geo and proxy checks
Risk lists
Payment pattern analysis
Address validation (PO box, freight forwarder)
ERP and fraud API signals
Auto-release safe orders; route risky orders to review

Benefits

Fewer chargebacks and false declines
Faster clean-order flow

Results

+2 to +4 percentage points approval rate
-15 to -30% chargebacks
-10 to -20% false declines

Scope

Owned rules, integrations, playbooks, dashboards, and weekly reviews

Links

Full case study: https://www.alexgranovskiy.com/case-study-order-risk-rules-engine/
More case studies: https://www.alexgranovskiy.com/tag/case-studies/