Lucy AI is a modernized, full-stack machine learning ecosystem designed for real-time market analysis and intent classification. Originally a legacy Python 2 system, it has been re-engineered into a high-performance FastAPI service featuring a robust SVM Pipeline and a dynamic React/amCharts 5 frontend.
- Intent Intelligence: A modernized Scikit-Learn SVC pipeline that classifies user queries with 86% accuracy and provides real-time confidence scores.
- Predictive Market Insights: Automagically identifies bullish/bearish trends using historical stock data via a custom Lucy Brain logic bridge.
- Modernized Infrastructure: Successfully migrated from legacy pickle formats to efficient joblib pipelines, ensuring Python 3.10+ compatibility.
- Dynamic Visualization: High-fidelity financial charts powered by amCharts 5, featuring real-time data streaming and responsive "Insight Overlays."
- Scalable Routing: An intelligent Agent Router that handles Web3-ready requests and balances model predictions with probability-based guardrails.
- FastAPI: High-performance asynchronous API framework.
- SQLAlchemy: ORM for robust data persistence and historical trend analysis.
- Scikit-Learn: Feature engineering (TF-IDF equivalent) and Linear SVM classification.
- Joblib: Optimized model serialization for fast cold-starts.
- React + TypeScript: Type-safe UI components for mission-critical reliability.
- amCharts 5: Advanced data visualization for complex time-series data.
- Tailwind CSS: Modern, responsive styling with glassmorphic UI elements.
Lucy was evaluated using a 10-fold Stratified Cross-Validation to ensure reliability across imbalanced datasets.
| Metric | Class 0 (Closed) | Class 1 (Open) | Combined | | Precision | 0.56 | 0.91 | 0.86 (Weighted) | | Recall | 0.52 | 0.92 | 0.86 (Weighted) | | F1-Score | 0.54 | 0.92 | 0.86 (Weighted) |
The model utilizes class_weight='balanced' to ensure the minority "Closed Question" class is handled with maximum sensitivity.
├── lucy/ # Legacy Feature Engineering Bridge (Modernized) ├── models/ # Serialized Joblib Pipelines & Vocabularies ├── routers/ # FastAPI Agent Logic & Insight Endpoints ├── data/ # Feature-engineered training sets ├── buildmodel.py # ML Pipeline training & Balancing logic └── evalmodel.py # ROC Curve & Classification performance scripts
- Clone & Install Dependencies
pip install -r requirements.txt- Seed the Market Brain
python seed_data.py- Launch the Agent
uvicorn main:app --reloadThis project demonstrates expertise in Legacy Code Modernization, MLOps (Model Deployment), and Full-Stack Financial Dashboarding.
Available for:
- Machine Learning Engineering roles
- Full-Stack AI Development
- Custom Trading Bot / Dashboard consultations