Track the sentiment. Measure the impact. Understand the markets.
FinanceWatcher is an AI-powered utility that analyzes financial news and social media for sentiment and short-term market impact. By combining LLM-powered RAG (Retrieval-Augmented Generation), sentiment scoring, and price trend tracking, it helps users gain actionable insights into how the market reacts to news.
-
News Ingestion & Embedding
Ingest financial news, generate vector embeddings using BAAI/bge-m3, and store metadata-rich entries in a ChromaDB vector store. -
Sentiment Analysis
Extract sentiment scoring provided by the MarketAux API. -
Impact Measurement
Quantify how a stock's price changed due to the news article. -
Model Context Protocol (MCP) Tools
Structured tools to let the LLM retrieve sentiment trends, calculate impact, price trends and current price. -
RAG-Powered Q&A
Use embeddings + metadata for context-aware LLM interactions — explore recent news with a smart query system that is semnatically aware of the company and other keywords, eg: 'BLK earnings', "Microsoft AI news" etc.
| Layer | Tools Used |
|---|---|
| Backend | Flask, Python |
| Embedding | BAAI/bge-m3 |
| Reranker | BGEReranker |
| LLM | LLama3 |
| Vector DB | ChromaDB |
| MCP Tools | Anthropic Python SDK (MCP protocol) |
| Data | MarketAux API |
| Frontend | React |
- Set up news ingestion pipeline
- Integrate Huggingface embeddings
- Implement sentiment scoring
- Calculate price-based impact scores
- Create core MCP tools
- Add basic RAG query flow
- Build frontend dashboard
FinanceWatcher by [Andrei Nanescu]