FinHub Portfolio — RAG + Research

A lightweight setup to ingest news, build a FAISS vector store, retrieve context via RAG, and (optionally) run a multi-agent research + paper-trading workflow.

Quick Start

1. Install dependencies

C:/Users/Admin/Desktop/Projects/FinHubPortfolio/.venv/Scripts/python.exe -m pip install -U -r requirements.txt

2. Ingest news for a ticker (creates rawdata/<TICKER>_articles_YYYYMMDD.csv)

C:/Users/Admin/Desktop/Projects/FinHubPortfolio/.venv/Scripts/python.exe "scripts/ingest_news.py" AAPL --months_back 3 --min_articles 60

3. Run a RAG query (optionally synthesize an answer with an LLM)

# Context only
C:/Users/Admin/Desktop/Projects/FinHubPortfolio/.venv/Scripts/python.exe "scripts/rag_query.py" AAPL "market risks and regulation" --out output_context.md

# With LLM synthesis (requires OPENAI_API_KEY)
$env:OPENAI_API_KEY = "<your_api_key>"
C:/Users/Admin/Desktop/Projects/FinHubPortfolio/.venv/Scripts/python.exe "scripts/rag_query.py" AAPL "How do current risks affect pricing power?" --llm gpt-4o-mini

4. Multi-agent research flow (optional)

# Runs ingestion → quality → features → model → risk → policy → paper execution → narrative
C:/Users/Admin/Desktop/Projects/FinHubPortfolio/.venv/Scripts/python.exe "scripts/run_research.py"

Layout

• src/rag/retriever.py: RAG retriever over local news using FAISS + sentence-transformers
• scripts/ingest_news.py: Ingest Yahoo Finance articles into rawdata/
• scripts/rag_query.py: Retrieve top-k context and optionally synthesize an answer
• scripts/run_research.py: Bridge into CrewAI research flow (kept separate)
• Multi Agent Assistant/crew_projects/: Existing multi-agent flows/tools (optional)

Notes

• Embeddings use langchain-huggingface to avoid deprecation warnings.
• For LLM synthesis, install langchain-openai and set OPENAI_API_KEY.
• You can safely use RAG without any multi-agent components.