I build intelligent systems.
I am an AI Engineer (an umbrella term for a machine learning engineer and generative AI engineer).
Currently building an AI harness at Devsinc and working towards becoming a Mechanistic Interpretability Researcher
Production AI features inside client products: retrieval services, LLM workflows, and the evaluation plumbing that keeps them shippable.
Owned features end to end — model, API, interface. Most model problems turned out to be data problems wearing a costume.
Long-form write-ups on fine-tuning, retrieval, state-space models and Linux internals. Everything reproducible.
A retrieval-augmented pipeline over clinical text: chunking that respects document structure, embedding choice tested against real queries, and answers traceable back to source.
Fine-tuned BLIP-2 on Flickr8k with LoRA adapters — data preparation, adapter placement, and what the loss curve actually told me about capacity.
A transformer built up from pseudo-code to C++: attention, positional encoding and the training loop written out by hand rather than imported.