COBOL Modernization Pipeline

Inspiration

I was inspired by the work and legacy of Grace Hopper, a pioneer in computer science and robotics who in 1959 had participated in the CODASYL consortium, helping to create the machine-independent programming language called COBOL.

What it does

This hackathon project proposes a secure, low-cost, and scalable AI-assisted pipeline for modernizing legacy COBOL systems, with a strong focus on understanding, planning, and collaboration rather than blind code translation. The system ingests COBOL source code and existing documentation, analyzes system structure and dependencies, and automatically generates high-level architectural views, modernization insights, and actionable plans. Instead of acting as a generic chatbot, it uses a template of specialized local AI agents to reason about code structure, architecture, risk, and modernization strategies.

A key feature is a collaborative UI that brings together COBOL code, modernization documentation, visual system diagrams, and generated issues in one workspace. Teams can explore the system at a high level, drill into specific modules, and see how risks, dependencies, and modernization tasks connect. The platform can automatically generate issues and plans of action tailored to constraints like budget, timeline, and target architecture, enabling coordinated execution across architects, engineers, and project managers.

The design is security-first and cost-aware: it runs locally or in controlled environments, avoids sending proprietary code to external services by default, and scales horizontally using stateless agents and cached analysis. Optional use of the Wolfram Language adds rigorous data analysis, dependency graphing, and quantitative risk or timeline modeling. Overall, the project positions itself as an “AI modernization architect”—helping organizations safely understand and plan COBOL modernization before making irreversible decisions.

How I built it

The system was built using Python as the core backend to orchestrate COBOL parsing, document ingestion, and an agent-based AI workflow, with ANTLR-based COBOL grammars extracting structural metadata that is embedded using local embedding models and stored in a vector database (e.g., FAISS/Chroma). Local LLMs coordinate through an agent framework to perform code understanding, architecture synthesis, and issue generation, while Wolfram Language is used for dependency graph analysis and quantitative metrics. A FastAPI layer exposes these capabilities to a React-based web UI, where Mermaid.js/D3.js render architecture and dependency visualizations. All components are containerized with Docker, communicate through internal REST APIs, and are designed to run on-prem or in a VPC with encrypted storage, cached embeddings, and stateless workers to keep the system secure, scalable, and cost-efficient. I used Google AI Studio for a UI demo of the project.

Challenges I ran into / What I learned

This project has a lot of working parts, so trying to implement everything and get it working together is even still a work in progress.

Accomplishments that I'm proud of

Getting to see this idea go from idea to plan of action to GitHub repo and UI demo: this is actually possible!

What's next for COBOL Modernization Pipeline

Developing everything! There is a ton of more code to write, COBOL code to test it out especially in testing the feasibility of scaling this project, and even more. Stay tuned in for more :)

Built With

• antlr
• cobol
• d3.js
• docker
• fastapi
• langchain
• mermaid
• python
• react
• typescript
• wolfram-technologies