How dodji1/julia works
Vitalia is an early-stage, source-grounded legal research assistant, initially focused on Benin tax law. The README documents a planned question-to-cited-answer workflow and corpus ingestion architecture, but the file tree contains no backend or frontend source implementing them; those components are shown as documented plans with null paths. The current repository evidence is limited to the README, the corpus manifest, and an HTML file whose role is not established here.
Main entry points: corpus_manifest.yaml Corpus manifest · raw Legal documents
Components 17
Legal research
- Frontend
- FastAPI
RAG workflow
- RAG orchestrator
- Query analyzer
- Retriever
- Reranker
- Context builder
- LLM provider
- DeepSeek
- Citation verifier
Corpus ingestion
- Embedder
- Extraction and OCR
- Article parser
Corpus and storage
- Corpus manifest
data/metadata/corpus_manifest.yaml - Passage store
- Legal documents
data/raw
Other components
- Researcher
Connections 17
- Researcher → Frontend: asks question
README.md - Frontend → FastAPI: sends query
README.md - FastAPI → RAG orchestrator: forwards query
README.md - RAG orchestrator → Query analyzer: analyzes query
README.md - RAG orchestrator → Retriever: retrieves passages
README.md - RAG orchestrator → Reranker: reranks passages
README.md - RAG orchestrator → Context builder: builds context
README.md - RAG orchestrator → LLM provider: requests answer
README.md - LLM provider → DeepSeek: calls provider
README.md - RAG orchestrator → Citation verifier: verifies citations
README.md - Retriever → Passage store: searches passages
README.md - Citation verifier → Passage store: reloads excerpts
README.md - Legal documents → Extraction and OCR: provides documents
README.md - Corpus manifest → Extraction and OCR: describes corpus
README.md - Extraction and OCR → Article parser: extracts text
README.md - Article parser → Embedder: provides passages
- Embedder → Passage store: stores embeddings
README.md