vkauhanau

 - End to end pipeline of Creation, training and implementing of ML models:
EDA, cleaning, and feature selection of noisy financial time series data 
Designed custom target engineering 
Combinating of weighted MAE + cross-entropy hybrid losses with class/timeframe balancing for evaluation of model
Trained multi-branch neural networks for multi-timeframe forecasting with PyTorch, combining weighted MAE + cross-entropy with class/timeframe balancing 
Implemented and tested trading strategies based on ML model with Backtrader library 
Developed a production inference service with REST endpoints for OHLCV ingestion, preprocessing, and real-time predictions 
- Created and integrated LLM-based assistant CoFo AI
Developed prompting strategies and structured schemas for market summarization and classification 
Integrated LLM with trading indicators and analytics tools, ensuring grounding and reducing hallucinations 
Implemented evaluation pipelines for LLM outputs via LLM as a judge strategy (correctness, structure, security) 
Fine-tuned interaction with external APIs and tools and optimized prompts for multilingual outputs 
Developed multi-branch market analysis pipeline