LegacyLens

LegacyLens is a RAG system for legacy COBOL codebases with citation-grounded answers, dependency mapping, and explicit fallback UX.

Live Demo

• App: <DEPLOYED_URL>
• Video: <DEMO_VIDEO_URL>

Quick Start

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

Full Setup

• See docs/SETUP.md for env vars, local Qdrant setup, ingest commands, and validation gates.

Architecture Overview

• See docs/ARCHITECTURE.md for full flow.
• Key path:
  • syntax-aware chunking
  • OpenAI embeddings (text-embedding-3-small)
  • Qdrant semantic retrieval + keyword fallback
  • query guardrails (vague input rejection, empty-index rejection, weak-evidence abstain)
  • answer generation + citations-only fallback
  • SSE streaming endpoint (POST /query/stream)

Query Examples (Required Scenarios)

1. Main entry point: Where is the main entry point?
2. CUSTOMER-RECORD modifiers: Where is CUSTOMER-RECORD modified?
3. CALCULATE-INTEREST explanation: Explain CALCULATE-INTEREST flow.
4. File I/O operations: Where is file I/O handled?
5. MODULE-X dependencies: What depends on MODULE-X?
6. Error handling patterns: Show error handling patterns.

Tech Stack

• FastAPI
• Qdrant
• OpenAI embeddings (text-embedding-3-small by default)
• OpenAI LLM answer generation
• COBOL corpus under data/ and src/legacylens/sample_codebase/

Deployment Link

• <DEPLOYED_URL>

Testing

pytest -q
python scripts/validate_corpus.py --codebase data
python scripts/validate_traceability.py