DOGE Lense: RAG-Powered Legacy Code Intelligence
AI-powered legacy code intelligence system that makes 850,000+ lines of COBOL and Fortran queryable through natural language, using a RAG pipeline with Voyage Code 2 embeddings, Qdrant vector search, and GPT-4o.
DOGE Lense: RAG-Powered Legacy Code Intelligence
Role: AI Engineer
Program: Gauntlet AI — 2-Month Immersive for AI Engineers
Live Demo: frontend-pied-alpha-71.vercel.app
GitHub: github.com/alediez2048/Gauntlet-Assignment-3
Tools: Python, FastAPI, Next.js 14, Voyage Code 2, Qdrant, GPT-4o, Cohere, Docker
Overview
DOGE Lense (LegacyLens) is a RAG-powered legacy code intelligence system that makes 850,000+ lines of COBOL and Fortran queryable through natural language. Users ask questions in plain English and receive cited answers with exact file:line references.
Eight Specialized Analysis Modes
- Code Explanation — Plain-English summaries with citations
- Dependency Mapping — Trace PERFORM/CALL chains across files
- Pattern Detection — Identify patterns across codebases
- Impact Analysis — "What breaks if this changes?"
- Documentation Generation — Auto-generate docs for undocumented code
- Translation Hints — Suggestions for modernizing to Python
- Bug Pattern Detection — Identify anti-patterns with severity levels
- Business Logic Extraction — Extract rules in plain English
Codebases Indexed
GnuCOBOL, GNU Fortran, LAPACK, BLAS, and OpenCOBOL Contrib.
Tech Stack
| Layer | Stack |
|---|---|
| Backend | Python 3.11, FastAPI, Click + Rich CLI |
| AI/ML | Voyage Code 2 (embeddings), GPT-4o (generation), Cohere (reranking) |
| Vector DB | Qdrant Cloud |
| Frontend | Next.js 14 |
| Deployment | API on Render (Docker), Web UI on Vercel |
