RepoMap: Instant Architecture Maps

Problem Onboarding to an unfamiliar codebase is slow and brittle. Engineers, reviewers, and students open a GitHub repo and face hundreds of files with no map of modules, routes, data paths, or how to run it. Docs go stale. Chatbots invent structure. Solution RepoMap turns a public GitHub URL into an interactive architecture dashboard. The user pastes a repo link; a Python FastAPI scanner shallow-clones the repo, walks the tree, and returns one deterministic JSON contract. A Next.js dashboard streams progress and renders eight views: Overview, Architecture, Data flow, Endpoints, Dependencies, Entry points, Key paths, and Health. Diagrams are Mermaid graphs you can inspect not a one-shot chat reply. How it works 1. Paste a public github.com/owner/repo URL. 2. The scanner clones (shallow, time-boxed), builds overview and structure, and when an AI key is set , asks a model for a layered architecture chart, then drops any invented paths that are not in the clone. 3. For Spring Boot Java, AST analysis (javalang) extracts entities, relations, endpoints, and risk findings (scope gaps, N+1, broken relations) with a scored risk level. 4. Multi-language structure fills endpoints, ERDs, import graphs, deps, entry points, and health signals across common stacks (Express, FastAPI, Go, Rails, Next.js, and more). 5. Results persist under scans/ so refresh keeps the demo ready. Target users - Engineers joining a new service or reviewing a PR across an unfamiliar tree - Students and mentors who need a map before diving into source - Hackathon and open-source contributors who must orient fast - Tech leads who want a quick health and structure snapshot of a public repo What makes it unique RepoMap is not “ask an LLM to summarize the repo.” Structure comes first from the filesystem and parsers; AI only upgrades diagrams and is validated against real paths. The JSON contract is stable and sortable, so the same tree yields the same baseline. Spring Boot gets deeper security/performance findings; every public repo still gets overview, architecture, deps, entry points, and health. Streaming status (clone → scan → overview → diagram) keeps the wait honest. Efficiency Shallow clone, file caps, module-aware scanning for large Java trees, and path validation keep maps fast and grounded. Users get a usable map in one paste no local setup of the target repo, no manual Mermaid authoring.