Software engineer working on machine learning pipelines, LLM desktop applications, and high-throughput data systems.
- FrameWords — Desktop video and audio analysis application built with Electron, React, Node.js, and Python. Combines serverless GPU pipelines (Modal L4/A100 running
faster-whisperand Qwen-VL) with local GGUF inference (node-llama-cpp) and multi-provider cloud LLM APIs. - opencl-quant-engine — GPU-accelerated backtesting and parameter sweep engine using custom OpenCL C99
fp64kernels and PyOpenCL, verified for bit-exact parity against Python/NumPy reference implementations. - clickhouse-realtime-grid-engine — Asyncio WebSocket ingestion and strategy evaluation engine backed by ClickHouse and Redis, with genetic algorithm parameter optimization and a web monitoring interface.
- mtf-lightgbm-optuna-pipeline — Multi-timeframe (1H/4H/1D) feature engineering, Optuna TPE hyperparameter optimization, and LightGBM / PyTorch LSTM training pipeline.
- orderbook-ml-scanner — Multi-exchange (Binance, Bybit, OKX) real-time L2 order book slippage estimator and Random Forest / Gradient Boosting signal scanner with a PyQt5 interface.
- Languages: Python, TypeScript, JavaScript, SQL, C (OpenCL C99)
- ML & AI: LightGBM, Scikit-learn, PyTorch, Optuna, Numba,
faster-whisper,node-llama-cpp, OpenAI / Anthropic / Gemini APIs - Data & Infrastructure: ClickHouse, DuckDB, Redis, PostgreSQL/Supabase, Polars, PyOpenCL, Asyncio, Docker, Modal
- Frontend & Desktop: Electron, React, Tailwind CSS, FastAPI, Flask, PyQt5/PyQt6
- Website: framewords.ai
- LinkedIn: linkedin.com/in/youssefdakkak
- Email: yodakkak@gmail.com