has the Fortran code and also a tool to help understand it, which I have not tried.
LegacyLens + NSHMP Fortran
This repository contains two things:
The original National Seismic Hazard Mapping Project (NSHMP) Fortran codebase and data/config assets.
LegacyLens, a retrieval-augmented code intelligence app built to analyze this legacy Fortran system with line-level evidence and architecture context.
Problem / Solution / What / Why / How
Problem
Legacy scientific Fortran systems are hard to work with:
critical logic is spread across many files and decades of style drift
call chains and dependencies are expensive to trace manually
onboarding often takes months before engineers can make safe changes
For teams, this means codebase insight that traditionally takes years of accumulated context now blocks delivery speed and change safety.
Solution
LegacyLens combines vector retrieval, lexical symbol search, and graph retrieval into one deterministic workflow so engineers can move from question to grounded evidence quickly.
What
LegacyLens is a code intelligence tool for NSHMP-style legacy repositories that returns:
architecture context (entrypoints, flows, impact)
line-level evidence and citations
retrieval diagnostics and confidence signals
Why
The goal is to compress understanding time:
from slow manual archaeology to fast, evidence-backed exploration
from uncertain edits to traceable, citation-grounded decisions
from siloed expert knowledge to team-accessible understanding
In practice: what used to take weeks or months to trace can often be scoped in hours.
How
Route each query to the best retrieval plan (vector, keyword+vector, graph+vector, or escalation).
Use graph signals to identify high-probability candidate files.
Retrieve and rerank Pinecone chunks with deterministic scoring and dedupe.
Expand context and enforce identifier guardrails.
Return architecture + evidence with debug telemetry.
What LegacyLens does
LegacyLens helps engineers query a large legacy Fortran repository with:
semantic retrieval from Pinecone
exact-identifier lexical retrieval
graph-assisted architecture retrieval via GitNexus MCP
deterministic reranking and grounded citations
a hybrid UI showing architecture graph + evidence panel