Things I've built and shipped.
Projects that went past prototype — focused on AI automation, voice, and the infrastructure layer that makes agent workflows actually reliable.
-
Automation
AI Workflow Automation
End-to-end pipelines for triage, synthesis, and delivery with multi-model orchestration. Designed for real operational use: resumable, recoverable, and observable — not just demo-complete.
Key elements: multi-agent routing, incident detection, automated digest delivery, and configurable escalation logic.
TypeScript Python LLM APIs -
Voice
Bada — Voice & Transcription
On-device speech-to-text app for macOS built on Apple Silicon. Uses local quantized models for low latency without a network round-trip. Designed for everyday dictation and voice note capture.
Native Swift UI, mlx inference backend, configurable language targets including French and Korean.
Swift MLX macOS Apple Silicon -
Research
Research & Agent Infrastructure
Tooling for AI-assisted systematic reviews and large-scale document screening. Resumable per-paper runners, baseline benchmarking against human gold sets, and multi-model eval pipelines.
Built to handle real research datasets — thousands of papers, multiple annotators, honest accuracy tracking.
Python Gemini API Data pipelines
Want to collaborate?
If you're building at the intersection of AI and operational workflows — or need someone to turn a working proof-of-concept into something production-grade — get in touch.
Contact →