Peter Bowman-Davis

OpenAI

Model training over hardware engineering

I partner with companies building technology that pertains to life, death, autonomy, and composability.

In varying degrees of completion:

• GameNgen Open Source Repro.
• LLM Numeric Bias Investigation.
• CommaVQ Open Source Repro.
• Physics-Informed-Neural-Network Demos.
• PreImage - CLI tool for Python dependency optimization.
• EdgeEmbed - Embedding models run at the network edge for lowest-latency vector search.
• JSGrad - Micrograd rewritten into Javascript for edge-ML.
• This website.

• World models, neural simulation, and model-based RL.
• National security, defense technology, and OSINT.
• Lightweight ML: edge inference, model quantization, digital signals processing.
• Controls, planning, and mechatronics engineering.
• Industrial, decorative, and interior design (Ceramics, Metals, Bauhaus, Dieter Rams).