I build production-minded AI systems - safety middleware, developer tools, and the infrastructure around them.
|
A Claude Code skill that fights the learn-nothing problem in AI-assisted coding. Auto-generates structured deep-dives on AI-written code — why it was written that way, the CS concepts in play, and where to go deeper. Works across Python, TypeScript, Go, and Rust.
|
Real-time safety & quality middleware for LLM apps. A FastAPI proxy intercepting every request/response across a 7-evaluator concurrent pipeline — prompt injection, PII, toxicity, hallucination, relevance. Drop-in for OpenAI, Anthropic, Gemini, or Ollama. Redis caching, fail-open guarantees, live dashboard.
|
|
NeurIPS 2023 optimizer reimplemented from scratch in PyTorch. Reached 88.26% on FashionMNIST, beating an AdamW baseline with lower memory cost from its sign-based update rule.
|
Adaptive AI learning platform with a terminal-style interface — adaptive quizzes, real-time progress tracking, minimalist dashboard.
|
Selected contributor building the Focus Group Management & Analysis Module — a new study type adding real-time multi-participant data capture, NLP-assisted thematic analysis, and sentiment scoring to the RUXAILAB platform. Mentored by Marc.
Vue 3 · Firebase · Python · scikit-learn · LiveKit
