How vedantkarne/aura-financial-intelligence works
AURA is an earnings-intelligence product for transcript Q&A, KPI lookup, and investment briefs. Users ask through the chat interface; the agent dispatches search, KPI, or report tools, which combine hybrid transcript retrieval and structured financial data before returning cited answers. Transcript ingestion and evaluation/monitoring are additional capabilities. The graph follows documented and indexed relationships; several unsampled modules are represented by their exact source paths, while uncertain internal calls are omitted.
Main entry points: qa_chain.py RAG answer chain · pipeline.py Ingestion pipeline · tools.py Agent tools · server.py FastAPI service
Components 25
User experience
- Next.js interface
frontend/src/app/page.tsx - Run monitor
frontend/src/app/monitor/page.tsx - FastAPI service
src/api/server.py - Streamlit interface
src/ui/app.py
Agent workflow
- Agent orchestrator
src/agents/orchestrator.py - Agent tools
src/agents/tools.py - Agent events
src/agents/events.py
Retrieval and generation
- RAG answer chain
src/generation/qa_chain.py - Answer prompts
src/generation/prompts.py - Query router
src/retrieval/router.py - Query transformer
src/retrieval/query_transformer.py - Hybrid retrieval
src/retrieval/hybrid_retriever.py - BM25 search
src/retrieval/bm25_retriever.py - Cross-encoder reranker
src/retrieval/reranker.py
Earnings data
- Vector store
src/retrieval/vector_store.py - KPI schema
src/extraction/schema.py - KPI extraction
src/extraction/kpi_extractor.py - Transcript corpus
raw_data/Earning_Call_Transcripts/cleaned_ECTs_dataset - Transcript parser
src/ingestion/file_parser.py - Chunking
src/ingestion/chunker.py - Ingestion pipeline
src/ingestion/pipeline.py - Embedding model
src/ingestion/embedder.py
Quality and monitoring
- Evaluation metrics
evaluation/compute_metrics.py - RAGAS evaluation
src/evaluation/ragas_eval.py
Other components
- Analyst
Connections 27
- Analyst → Next.js interface: asks questions
README.md - Next.js interface → FastAPI service: sends requests
- FastAPI service → Agent orchestrator: invokes agent
src/api/server.py - FastAPI service → Agent tools: serves tools
src/api/server.py - Agent orchestrator → Agent tools: dispatches tools
src/agents/orchestrator.py - Agent orchestrator → Agent events: emits events
src/agents/orchestrator.py - Run monitor → Agent events: displays events
- Agent tools → RAG answer chain: searches transcripts
src/agents/tools.py - Agent tools → KPI schema: queries KPIs
src/agents/tools.py - Agent tools → RAG answer chain: builds report evidence
src/agents/tools.py - RAG answer chain → Query router: routes queries
src/generation/qa_chain.py - RAG answer chain → Query transformer: rewrites queries
src/generation/qa_chain.py - RAG answer chain → Hybrid retrieval: retrieves candidates
src/generation/qa_chain.py - Hybrid retrieval → Vector store: searches vectors
src/retrieval/hybrid_retriever.py - Hybrid retrieval → BM25 search: searches keywords
src/retrieval/hybrid_retriever.py - RAG answer chain → Cross-encoder reranker: reranks candidates
src/generation/qa_chain.py - RAG answer chain → Answer prompts: formats answers
src/generation/qa_chain.py - Ingestion pipeline → Transcript corpus: discovers files
src/ingestion/pipeline.py - Ingestion pipeline → Transcript parser: parses metadata
src/ingestion/pipeline.py - Ingestion pipeline → Chunking: chunks text
src/ingestion/pipeline.py - Ingestion pipeline → Vector store: indexes chunks
src/ingestion/pipeline.py - Vector store → Embedding model: embeds documents
src/retrieval/vector_store.py - Ingestion pipeline → KPI extraction: extracts metrics
src/ingestion/pipeline.py - KPI extraction → KPI schema: creates records
src/extraction/kpi_extractor.py - Analyst → Streamlit interface: uses chat
- RAGAS evaluation → RAG answer chain: evaluates answers
src/evaluation/ragas_eval.py - Evaluation metrics → RAGAS evaluation: reports quality