How wajih1234/legaltun works
LexTN is a French-language JORT legal-question chatbot with a React interface and two retrieval approaches. The shown frontend sends questions to the classical RAG endpoint; that API delegates retrieval and answer generation, returning answers and source chunks. A separate Flask GraphRAG API generates read-only Cypher and summarizes Neo4j results. The README also documents PDF ingestion into Qdrant and knowledge-graph construction. Those corpus-building links are documented, but their runtime details were not sampled; the UI’s shown endpoint does not select GraphRAG.
Main entry points: ChatArea.jsx Chat workflow · app.py Graph query API · neo4j_client.py Neo4j connection · App.jsx Application shell
Components 26
Chat interface
- Application shell
legal-chat-front/src/App.jsx - Chat workflow
legal-chat-front/src/components/ChatArea.jsx - Question input
legal-chat-front/src/components/InputBar.jsx - Answer display
legal-chat-front/src/components/MessageBubble.jsx - Conversation sidebar
legal-chat-front/src/components/Sidebar.jsx - Conversation state
legal-chat-front/src/hooks/useConversations.js - API client
legal-chat-front/src/services/api.js
Classical RAG
- RAG API
ragclassi/app.py - RAG orchestration
ragclassi/pipeline.py - Vector retrieval
ragclassi/retriever.py - Qdrant collection
ragclassi/qdrant_storage - Answer generation
ragclassi/generator.py
GraphRAG
- Graph query API
back/app.py - Entity resolution
back/entity_extractor.py - Cypher generation
back/llm_cypher.py - Schema discovery
back/schema_fetcher.py - Query validation and execution
back/cypher_runner.py - Neo4j connection
back/neo4j_client.py - Result summarization
back/answer_generator.py
Corpus preparation
- JORT PDF corpus
ragclassi/docs - PDF extraction and chunking
ragclassi/document_loader.py - Embedding preparation
ragclassi/embd.py - Triplet extraction
back/entity_extractor.py
Other components
- Questioner
- LLM service
- Knowledge graph
Connections 33
- Questioner → Chat workflow: asks
legal-chat-front/src/components/ChatArea.jsx - Chat workflow → Question input: renders
legal-chat-front/src/components/ChatArea.jsx - Chat workflow → API client: submits question
legal-chat-front/src/components/ChatArea.jsx - API client → RAG API: HTTP POST
legal-chat-front/src/services/api.js - RAG API → RAG orchestration: calls
ragclassi/app.py - RAG orchestration → Vector retrieval: retrieves chunks
ragclassi/pipeline.py - RAG orchestration → Answer generation: generates answer
ragclassi/pipeline.py - Vector retrieval → Qdrant collection: searches vectors
ragclassi/retriever.py - Vector retrieval → Answer generation: returns chunks
ragclassi/pipeline.py - Answer generation → LLM service: generates response
README.md - Chat workflow → Answer display: renders messages
legal-chat-front/src/components/ChatArea.jsx - Application shell → Chat workflow: renders
legal-chat-front/src/App.jsx - Application shell → Conversation sidebar: renders
legal-chat-front/src/App.jsx - Application shell → Conversation state: uses state
legal-chat-front/src/App.jsx - Conversation sidebar → Conversation state: manages conversations
legal-chat-front/src/App.jsx - Chat workflow → Conversation state: updates messages
legal-chat-front/src/App.jsx - Questioner → Graph query API: asks
- Graph query API → Entity resolution: resolves entities
back/app.py - Graph query API → Cypher generation: generates query
back/app.py - Graph query API → Query validation and execution: validates and runs
back/app.py - Graph query API → Result summarization: summarizes results
back/app.py - Cypher generation → Schema discovery: uses schema
back/llm_cypher.py - Schema discovery → Neo4j connection: queries through
back/schema_fetcher.py - Query validation and execution → Neo4j connection: runs queries through
back/cypher_runner.py - Neo4j connection → Knowledge graph: connects
- Entity resolution → Neo4j connection: resolves against
back/entity_extractor.py - Cypher generation → LLM service: uses LLM API
README.md - Result summarization → LLM service: uses LLM
- JORT PDF corpus → PDF extraction and chunking: provides documents
README.md - PDF extraction and chunking → Embedding preparation: chunks for embeddings
README.md - Embedding preparation → Qdrant collection: stores vectors
README.md - JORT PDF corpus → Triplet extraction: source corpus
README.md - Triplet extraction → Knowledge graph: builds graph
README.md