🚀 Shipped
SpaX
Pythonic, type-safe search space configuration for hyperparameter optimization, neural architecture search, and ML experiment tracking. Built to eliminate boilerplate and enforce best practices from research to production.
Features:
- Declarative search space definition with automatic inference
- Conditional parameters with complex dependency logic
- Nested and polymorphic configurations
- Native Optuna integration for HPO
- Iterative search space refinement based on results
- Multi-format serialization (JSON/YAML/TOML)
Stack: Python • Pydantic • Optuna
TickVault
High-performance financial tick data pipeline for Dukascopy Bank’s historical datafeed. Built for quantitative researchers and algorithmic traders who need reliable access to high-resolution market data.
Features:
- Concurrent downloads with intelligent resume capability
- Multi-proxy pipeline for distributed downloading
- Efficient decompression and decoding
- SQLite metadata tracking and gap detection
- Pandas and NumPy integration
Stack: Python • httpx • NumPy • Pandas • SQLite • LZMA
ProxyRotator
Async Python library for managing VMESS proxy and user-agent rotation with automatic subscription updates, connection testing, and stealth-focused user-agent selection. Built for resilient web scraping workflows.
Features:
- Automatic proxy rotation with subscription support
- Connection testing and filtering of working proxies
- User-agent rotation with globally popular patterns
- Rate limiting with jitter for natural request patterns
- Thread-safe with context manager support
Stack: Python • httpx • Xray-core • Pydantic • asyncio
🔬 In Development
Clean-TS
Modular, Pythonic reimplementation of canonical time-series architectures. Makes archaic, opaque TS codebases readable, extensible, and reproducible.
Status: Refactoring • ~1 month to release
Lightning HPO Playbooks
Industry-standard examples and guides for model training, optimization, and research using PyTorch Lightning. Covers SOTA practices for NAS, HPO, distributed training, and production-ready ML pipelines.
Status: Planning
Financial RL Environment
High-performance, parallelized Gymnasium environment for algorithmic trading research. Built for large-scale RL training with custom reward formulations.
Status: Planning