Open to opportunities

Stephen Diaz

@stephendiaz

Message

I build private, offline desktop AI systems powered by local LLMs.

Panama

Message

What I'm looking for

I'm looking to build privacy-preserving, local-first AI products where I can own architecture, model integration, benchmarking, and production delivery end to end.

I built SovNode, a fully offline PyQt6 desktop AI client that routes queries between a lightweight intent classifier and local reasoning models through Ollama. It gives users data sovereignty without relying on cloud AI providers or paid APIs.

I designed the system end to end, from local RAG with FAISS and AST-based code parsing to asynchronous inference, persistent session memory, speech-to-text with faster-whisper, and Windows packaging. I benchmarked sub-20ms intent routing, sustained 15.5–19.0 tokens per second on an AMD Radeon RX 5500 XT, and zero VRAM leakage across model swaps.

I also write about local AI systems architecture and build tactical AI, scripting, and memory-manipulation tools for game modding projects. I care about lean dependencies, privacy-preserving infrastructure, and clean, defensible production-quality code.

Experience

Work history, roles, and key accomplishments

Current

AI Systems Engineer

Current

Self Employed

Jan 2024 - Present (2 years 7 months)

Designed, built, and benchmarked SovNode end to end, including GPU acceleration tuning, memory profiling, and quantized local model evaluation. Maintained a lean dependency footprint and prioritized clean, production-quality code across personal projects.

Python Ollama Faiss GPU Acceleration Model Quantization REST APIs Git