KonveyN2AI: Unlocking the Business Logic Hidden in COBOL

Agent & Assistant
, , ,
1

Hey everyone,

I’ve been working with Neeharika Vemulapati (@vrneeharika7) on a project called KonveyN2AI for the Agent 4 Buildathon, and wanted to share what we built and what we learned.

image
image1362×1239 212 KB

We started with a simple problem.

Many important banking and payment systems still run on COBOL. The hard part in modernizing them is not just reading the code. It is understanding why the code was written that way in the first place.

You can reverse engineer the structure. But the business decisions behind years of logic are often missing. And that missing context is where migration risk shows up.

To tackle that, we built KonveyN2AI.

It parses COBOL programs into a semantic knowledge graph, extracts business rules, captures decision context, and measures migration readiness using a Decision Completeness Score.

We also built an AI Migration Advisor that operates on a structured ontology rather than relying solely on raw LLM output.

We built and deployed the project end-to-end on Replit using Replit Agent for iterative development across parsing logic, APIs, and UI. The backend is built with Python and FastAPI, with an interactive dashboard for visualization.

One insight stood out when we tested it on the AWS CardDemo portfolio.

Some programs looked technically ready to migrate, but had very low decision-context coverage. That means they may still carry serious hidden risks because the business logic behind them is not fully documented.

The hardest part was not parsing COBOL. It was finding and organizing the “why” behind business rules.

Our approach was to combine deterministic parsing with manual and inferred decision capture, and to keep structural understanding separate from contextual understanding.

You can explore it here:
https://konveyn2ai.replit.app/

Next, we want to support full project ingestion across multiple file systems, improve automated WHY inference, and make risk visualization better for larger portfolios.

Would love feedback, especially on:

  • decision-context capture

  • UI/UX clarity

  • real-world migration use cases

Also curious how others are dealing with undocumented business logic in legacy systems.

1 Like