How abhasbali/samsung-hackathon works
CodeFusion indexes repositories into semantic code snippets and answers natural-language questions with ranked, provenance-bearing results rather than generated prose. The main workflow is repository indexing (parse and chunk, construct representations and retrieval indexes), followed by intent-aware search, graph-aware retrieval and ranking. It also tracks incremental Git updates and snippet lineage across versions; the API and frontend expose search, index/version status, call chains and experiment results. Edges are limited to README statements and source-index references or excerpts; several implementation details are unsampled, so uncertain connections are omitted.
Main entry points: pipeline.py Retrieval engine · incremental.py Incremental indexing · App.tsx Web interface · chunker.py Parse and chunk
Components 34
Code indexing
- Parse and chunk
src/codefusion/parsing/chunker.py - Tree-sitter parser
src/codefusion/parsing/tree_sitter_parser.py - Parser fallback
src/codefusion/parsing/fallback.py - Identifier tokens
src/codefusion/parsing/identifiers.py - Code representations
src/codefusion/representations
Query and retrieval
- Retrieval engine
src/codefusion/retrieval/pipeline.py - Embedding providers
src/codefusion/retrieval/embeddings.py - Dense retrieval
src/codefusion/retrieval/dense.py - BM25 retrieval
src/codefusion/retrieval/bm25.py - Symbol retrieval
src/codefusion/retrieval/symbols.py - Call graph
src/codefusion/graph/builder.py - Graph scoring
src/codefusion/graph/scoring.py - Rank fusion
src/codefusion/retrieval/fusion.py - Reranking
src/codefusion/retrieval/reranker.py - Dedup and diversity
src/codefusion/retrieval/diversity.py - Snippet store
src/codefusion/retrieval/pipeline.py
Version history
- Incremental indexing
src/codefusion/versions/incremental.py - Git tracking
src/codefusion/versions/git_tracker.py - Snippet lineage
src/codefusion/versions/lineage.py
API and UI
- HTTP API
src/codefusion/api/main.py - Web interface
frontend/src/App.tsx - API client
frontend/src/api.ts - Search view
frontend/src/components/SearchView.tsx - Result details
frontend/src/components/ResultCard.tsx - Index and versions
frontend/src/components/IndexView.tsx - Experiments view
frontend/src/components/ExperimentsView.tsx
Evaluation
- Evaluation harness
src/codefusion/evaluation/official.py - MTEB encoder
src/codefusion/evaluation/mteb_encoder.py - Evaluation metrics
src/codefusion/evaluation/metrics.py - Run tracking
src/codefusion/evaluation/tracking.py - Experiment artifacts
src/codefusion/storage/artifacts.py
Other components
- Engineer
- Code repository
- Ranked snippets
Connections 39
- Engineer → Web interface: uses
- Web interface → API client: delegates
frontend/src/App.tsx - API client → HTTP API: HTTP requests
- Web interface → Search view: renders
frontend/src/App.tsx - Web interface → Index and versions: renders
frontend/src/App.tsx - Web interface → Experiments view: renders
frontend/src/App.tsx - Search view → API client: requests search
frontend/src/components/SearchView.tsx - Search view → Result details: renders results
frontend/src/components/SearchView.tsx - Index and versions → API client: requests indexing
frontend/src/components/IndexView.tsx - Experiments view → API client: loads experiments
frontend/src/components/ExperimentsView.tsx - Code repository → Incremental indexing: indexes versions
- Incremental indexing → Git tracking: reads history
src/codefusion/versions/incremental.py - Incremental indexing → Parse and chunk: chunks changes
src/codefusion/versions/incremental.py - Incremental indexing → Retrieval engine: adds snippets
src/codefusion/versions/incremental.py - Incremental indexing → Snippet lineage: matches lineage
src/codefusion/versions/incremental.py - Parse and chunk → Tree-sitter parser: uses parser
src/codefusion/parsing/chunker.py - Parse and chunk → Parser fallback: falls back
src/codefusion/parsing/chunker.py - Parse and chunk → Identifier tokens: tokenizes names
- Code repository → Parse and chunk: parses code
- Retrieval engine → Code representations: builds views
src/codefusion/retrieval/pipeline.py - Retrieval engine → Embedding providers: embeds snippets
src/codefusion/retrieval/pipeline.py - Retrieval engine → Dense retrieval: retrieves dense
src/codefusion/retrieval/pipeline.py - Retrieval engine → BM25 retrieval: retrieves BM25
src/codefusion/retrieval/pipeline.py - Retrieval engine → Symbol retrieval: retrieves symbols
src/codefusion/retrieval/pipeline.py - Retrieval engine → Call graph: expands candidates
src/codefusion/retrieval/pipeline.py - Retrieval engine → Graph scoring: scores structure
src/codefusion/retrieval/pipeline.py - Retrieval engine → Rank fusion: fuses rankings
src/codefusion/retrieval/pipeline.py - Retrieval engine → Reranking: reranks candidates
src/codefusion/retrieval/pipeline.py - Retrieval engine → Dedup and diversity: deduplicates results
src/codefusion/retrieval/pipeline.py - Retrieval engine → Snippet store: stores snippets
src/codefusion/retrieval/pipeline.py - HTTP API → Retrieval engine: searches
src/codefusion/api/main.py - HTTP API → Incremental indexing: indexes repository
src/codefusion/api/main.py - Retrieval engine → Ranked snippets: returns ranked snippets
- HTTP API → Ranked snippets: serves search output
src/codefusion/api/main.py - Evaluation harness → MTEB encoder: uses encoder
src/codefusion/evaluation/official.py - Evaluation harness → Evaluation metrics: computes metrics
src/codefusion/evaluation/official.py - Evaluation harness → Run tracking: tracks runs
src/codefusion/evaluation/official.py - Evaluation harness → Experiment artifacts: writes artifacts
- Experiments view → Experiment artifacts: displays results