How aerex0/recor-rag works
ReCOR RAG is a transcript-oriented conversational analytics project: it develops query datasets, runs retrieval-augmented answer generation, and presents answers with transcript evidence. The principal user workflow is to submit a curated dataset query through the web interface, run the answer-generation notebook against the RAG assets and model server, and inspect the answer and evidence. The README documents three query-generation approaches and evaluation; sampled code gives more detail for method 2 and the web chat state. Several notebook and backend relationships are documented only at a high level, so those links are kept broad.
Main entry points: main.py Pipeline orchestration · auto_evol.py Query evolution · answer-generation.ipynb Answer generation · self_instruct.py Self-Instruct
Components 21
Query creation
- Taxonomy generation
queries-generation/method-1/query-generation - Query evaluation
queries-generation/method-1/query-evaluation - Seed generation
queries-generation/method-2/seed_generator.py - Pipeline orchestration
queries-generation/method-2/main.py - Self-Instruct
queries-generation/method-2/self_instruct.py - Query evolution
queries-generation/method-2/auto_evol.py - Evolution checks
queries-generation/method-2/utils/failure_detector.py - Prompt templates
queries-generation/method-2/utils/prompts.py - Pipeline settings
queries-generation/method-2/config.py - Groq client
queries-generation/method-2/utils/llm_client.py - Transcript-based generation
queries-generation/method-3
RAG runtime
- Answer generation
recor-rag/answer-generation.ipynb - FAISS index
recor-rag/const/faiss_index - Transcript summaries
recor-rag/const/summaries-20k.json - vLLM model server
- Answer and evidence
recor-rag/answer-generation.ipynb
Web app
- Chat interface
website/app/page.tsx - Chat state
website/app/stores/useChatStore.ts - Chat API route
website/app/api/chat/route.ts
Data assets
- Curated queries
simulated-real-query-system
Other components
- Analyst
Connections 23
- Analyst → Chat interface: submits query
README.md - Chat interface → Chat state: uses chat state
website/app/page.tsx - Curated queries → Answer generation: provides test queries
README.md - Answer generation → Answer and evidence: produces answer
README.md - Answer generation → FAISS index: uses retrieval index
README.md - Answer generation → Transcript summaries: uses summaries
- Answer generation → vLLM model server: queries model
README.md - Pipeline orchestration → Self-Instruct: runs stage one
queries-generation/method-2/main.py - Pipeline orchestration → Query evolution: runs stage two
queries-generation/method-2/main.py - Pipeline orchestration → Groq client: creates clients
queries-generation/method-2/main.py - Pipeline orchestration → Pipeline settings: gets output paths
queries-generation/method-2/main.py - Self-Instruct → Groq client: requests completions
queries-generation/method-2/self_instruct.py - Self-Instruct → Prompt templates: uses prompts
queries-generation/method-2/self_instruct.py - Self-Instruct → Pipeline settings: uses settings
queries-generation/method-2/self_instruct.py - Query evolution → Groq client: requests completions
queries-generation/method-2/auto_evol.py - Query evolution → Evolution checks: checks evolutions
queries-generation/method-2/auto_evol.py - Query evolution → Prompt templates: uses prompts
queries-generation/method-2/auto_evol.py - Query evolution → Pipeline settings: uses settings
queries-generation/method-2/auto_evol.py - Seed generation → Groq client: requests completions
queries-generation/method-2/seed_generator.py - Taxonomy generation → Query evaluation: queries evaluated
- Taxonomy generation → Curated queries: contributes queries
- Transcript-based generation → Curated queries: contributes queries
- Pipeline orchestration → Curated queries: generates query data