KonveyN2AI — Mapping legacy COBOL systems into a knowledge graph (Agent 4 Buildathon)

Showcase
, ,
1

Hey everyone :waving_hand:

I’ve been working with Sai Nikhil Damacherla (@SaiNikhilNikhil) on a project called KonveyN2AI for the Agent 4 Buildathon, and wanted to share what we built and what we learned along the way.

We started with a simple observation:
modernizing legacy COBOL systems is risky—not because the code is unreadable, but because the reason behind the code is missing. While structure can be reverse-engineered, the business decisions behind decades of logic are often lost, and that missing context is what causes migration failures.

To address this, we built KonveyN2AI — a system that parses COBOL programs into a semantic knowledge graph, identifies embedded business rules, captures decision context (“WHY”), and evaluates migration readiness using a Decision Completeness Score. On top of this, we introduced an AI Migration Advisor that works on structured ontology rather than raw LLM output.

The project was built end-to-end on Replit, using Replit Agent to iteratively develop parsing logic, APIs, and UI components, and deployed as a live application. The backend is built with Python/FastAPI, with an interactive dashboard for visualization.

While testing on the AWS CardDemo portfolio, one key insight stood out: some programs appeared technically ready to migrate but had very low decision-context coverage — meaning they carry hidden risk due to undocumented business logic.

The biggest challenge wasn’t parsing COBOL — it was identifying and structuring the “WHY” behind business rules. We approached this by combining deterministic parsing with manual and inferred decision capture, and by separating structural understanding from contextual understanding.

You can explore the project here:
:backhand_index_pointing_right: https://konveyn2ai.replit.app/

Next steps include supporting full project ingestion (multi-file systems), improving automated WHY inference, and enhancing visualization of risk across large portfolios.

Would love feedback — especially around decision-context capture, UI/UX clarity, and real-world use cases.

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

Screenshot 2026-04-02 at 2.26.59 PM
Screenshot 2026-04-02 at 2.26.59 PM1920×1247 245 KB

1 Like
2

Looks like an awesome project!

Yes, that’s certainly one of the challenges, in particular because the required knowledge is often known only by very few people, often on the verge of retiring. And besides that, those systems tend to be in the center of a huge web of dependent systems, some of them neartime or realtime systems, that are required to be highly reliable. So you do not only need to know the system itself but usually have to have an intimate knowledge of the broader context. And the replacement might need significant architectural changes. …