Crypto AI-Tracker: Microservices Dashboard with Hybrid Sentiment Analysis
Retail investors bleed money due to market noise and emotional bias. This platform acts as a personal quant. It’s a microservices-based dashboard that fuses real-time portfolio tracking with an NLP-driven 'Fear & Greed' index. Zero guesswork, just data-backed insights.
Tech Stack
Docker & Docker Compose
Ensured full containerization. I isolated every microservice so the system runs identically across macOS, Linux, and Windows without environment conflicts.
Python (FastAPI) + PyTorch
The perfect engine for heavy ML lifting. FastAPI handles asynchronous requests while the dedicated inference service runs FinBERT, keeping the main API Gateway completely unblocked.
Node.js (Express)
Acts as the API Gateway and central orchestrator. It securely manages business logic, portfolio CRUD operations, and bridges the React client with the AI engine.
React + Vite
Chosen for high-performance SPA rendering. Ensures the dashboard UI and heavy data visualizations update instantly without layout thrashing.
Features
- Two-stage Hybrid AI pipeline (FinBERT + Gemini) generating hallucination-free market insights.
- Real-time cryptocurrency portfolio tracking with dynamic data visualization.
- Automated news aggregation and processing from multiple financial sources.
- Decoupled microservices architecture ensuring high scalability and fault tolerance.
Results
- Architected a novel two-stage Hybrid AI pipeline, effectively eliminating the risk of LLM hallucinations in financial sentiment analysis.
- Built a resilient microservices ecosystem with secure internal networking between Node.js and Python containers.
- Separated system logic into the 'Quantitative Analyst' (scoring) and 'Qualitative Reporter' (summary), achieving enterprise-grade objectivity.
Challenges & Solutions
- Challenge
- AI Hallucinations: Generative LLMs often fabricate facts or misjudge the sentiment of financial news, which is catastrophic for trading decisions.
- Solution
- Engineered a hybrid pipeline. First, a local FinBERT model acts as the 'Ground Truth', delivering strict mathematical probability scores. Only then does Gemini receive these hard numbers alongside the news to draft a human-readable report. Gemini is constrained by math, making hallucinations impossible.
What I Learned
- Mastered cross-service communication and networking between Node.js and Python ecosystems via Docker.
- Learned how to effectively constrain Generative AI outputs using deterministic Machine Learning models.
- Gained practical experience in designing and orchestrating a scalable microservices architecture from scratch.