Local-first, privacy-preserving AI platform for real-time stock & crypto analysis, price prediction, and conversational financial intelligence — powered entirely by open-source LLMs.
A production-grade AI financial analysis system built on local infrastructure with zero cloud dependency. Combines Retrieval-Augmented Generation (RAG), Model Context Protocol (MCP) agentic orchestration, and multimodal market analysis to deliver institutional-grade financial intelligence on commodity hardware.
Built entirely in Python. Runs fully offline. No data leaves your machine.
┌─────────────────────────────────────────────────────────────────┐
│ User Interface │
│ Conversational Chat │ Analysis Dashboard │
└───────────────────┬─────────────────────────┬───────────────────┘
│ │
┌──────────▼──────────┐ ┌──────────▼──────────┐
│ Chat Interface │ │ Analysis Engine │
│ Llama 3.1:8b │ │ Qwen2.5:14b │
│ (Conversational) │ │ (Deep Analysis) │
└──────────┬──────────┘ └──────────┬──────────┘
│ │
┌──────────▼─────────────────────────▼──────────┐
│ MCP Orchestration Layer │
│ stock_mcp_server.py │ news_mcp_server.py │
└──────────┬─────────────────────────┬──────────┘
│ │
┌──────────▼──────────┐ ┌──────────▼──────────┐
│ RAG Pipeline │ │ Live Data Layer │
│ ChromaDB │ │ Stock APIs │
│ UAE-Large-V1 │ │ Crypto APIs │
│ angle_emb │ │ News Feeds │
└─────────────────────┘ └─────────────────────┘
│
┌──────────▼──────────┐
│ Ollama Runtime │
│ Local LLM Inference│
│ GPU/CPU Optimized │
└─────────────────────┘
- Real-time stock and cryptocurrency price tracking
- Candlestick pattern recognition with AI interpretation
- Technical indicator analysis (RSI, MACD, Bollinger Bands)
- Multimodal processing of financial charts and visual market data
- EOD (End-of-Day) price forecasting for stocks and crypto
- End-of-Week prediction models using historical data
- Trend confidence scoring and risk framing
- Multi-tool AI agents that autonomously fetch, analyze, and report
stock_mcp_server.py— live market data tool servernews_mcp_server.py— financial news aggregation and sentiment- Agent coordination for complex, multi-step financial queries
- Full natural language chat interface for ad-hoc financial queries
- Context-aware responses grounded in live and historical data
- Supports: "What's the outlook for NVDA this week?", "Summarize BTC sentiment from today's news"
- 100% local inference — no data transmitted to external APIs
- All embeddings, vectors, and models stored on local disk
- Suitable for sensitive financial data and compliance-conscious environments
| Model | Purpose | Parameters |
|---|---|---|
Qwen2.5:14b |
Deep financial analysis & prediction | 14B |
Qwen2.5-Coder:14b |
Code generation & data processing | 14B |
Llama 3.1:8b |
Conversational chat interface | 8B |
Embedding Model: UAE-Large-V1 via angle_emb
Vector Store: ChromaDB (persistent, local)
Inference Runtime: Ollama
Language Python
LLM Runtime Ollama
Models Qwen2.5:14b · Qwen2.5-Coder:14b · Llama3.1:8b
Embeddings UAE-Large-V1 (angle_emb)
Vector DB ChromaDB
Orchestration MCP (Model Context Protocol)
Data Sources Financial APIs · News Feeds · Historical OHLCV
Analysis Multimodal AI · Candlestick Pattern Recognition
Prediction RAG-augmented forecasting with historical context
| Decision | Rationale |
|---|---|
| Local-only inference | Data privacy, zero latency on API calls, cost elimination |
| Qwen2.5:14b for analysis | Superior financial reasoning vs smaller models; fits 16GB VRAM |
| Llama3.1:8b for chat | Optimized for conversational flow, faster response latency |
| UAE-Large-V1 embeddings | State-of-the-art semantic retrieval for financial text |
| ChromaDB | Persistent local vector store with simple Python integration |
| MCP for tool orchestration | Clean agent/tool separation; extensible to new data sources |
| Multimodal input | Candlestick charts carry information not captured in raw OHLCV |
Active development — personal research platform. Source code is proprietary. Architecture and design documentation shared for portfolio purposes.
Ahmed M. Eldeep Principal Full Stack Engineer & Cloud Architect linkedin.com/in/eldeep
© 2026 Ahmed M. Eldeep. All rights reserved. Source code is proprietary and not available for redistribution.