AI that ships — not slideware.
Most "AI solutions" are a thin wrapper around someone else's API and a demo that falls apart in production. I build the other kind: models trained on your data, running where you need them, doing work you can measure.
I've built and trained models from scratch — computer vision (YOLOv8, CUDA), fine-tuning with LoRA and DPO, local inference on Qwen and LM Studio, and agentic systems on Anthropic and OpenAI wired together with MCP. Proof-of-concept to a production system your team can actually run.
What I build
- Custom models, trained from scratch — vision, classification, prediction, and recommendation on your own data, not a generic endpoint.
- LLM applications & agents — retrieval, tool use, and multi-step agents connected to your real systems.
- Local & private inference — models that run on your hardware or in your VPC when the data can't leave the building.
- Intelligent automation — pipelines that read unstructured data, make a decision, and take action.
Why me
I don't hand off to a junior team. Concept, data, training, deployment, and the operator UI — one person accountable end to end. I've done it for a fielded RF-detection defense system and for fintech platforms running live market analysis.
If it can be built, I want to build it — end to end, one person accountable.