I build AI tools that are meant to be practical and run on normal hardware.

Most of my recent work focuses on making AI more accessible - whether that's language models that don't require a datacenter, RAG systems you can run locally or tools to check if your ML models might be biased.

PolyglotLite - Lightweight multilingual language models (100-500M params) for consumer hardware. Supports 50+ languages.

RAGKit - Simple framework for building document Q&A apps. Local-first, no API keys required for basic setup.

FairLens - Bias detection toolkit for ML models. Quick audits without the complexity of larger frameworks.

• Efficient ML - getting good results without massive compute
• Multilingual NLP - there's a lot of underserved languages out there
• Responsible AI - bias detection, fairness metrics, that sort of thing
• Making AI tools that regular developers can actually use

If you're working on something similar or have questions about any of these projects, feel free to open an issue or reach out.