Open Source • 2026

CodebaseQA

Codebase exploration with source-cited answers, guided learning and an interactive dependency graph, built around repository code instead of architecture guesses.

My Role

Designed the indexing, retrieval, learning and graph workflows and implemented them with AI assistance.

Team Size: 1 person

Problem

Getting oriented in an unfamiliar repository involves scattered documentation, repeated searches and uncertain assumptions about how the code connects.

Approach

Built repository indexing with Tree-sitter, intent-routed hybrid retrieval, caches and streamed answers. Citation checks compare generated references with indexed paths and file lengths. A deterministic import resolver builds the dependency graph, which React renders with capped views and layout fallback.

Impact

Combined code search, cited answers, learning tracks and graph exploration in one self-hostable application. Replacing the model-generated graph removed provider calls from the default graph path; chat still uses models. User adoption and onboarding-time improvements have not been measured.

Key Metrics

9 Tree-sitter grammars
Repository parsing
6 scoring profiles
Retrieval routing
0 LLM calls
Default dependency graph path

Technologies

PythonFastAPINext.jsReactTypeScriptTree-sitterChromaDBRedisReact Flow

Links