AI Product Engineer · LLM Systems · AI Security
Building the bridge between impressive AI demos and reliable production systems.
I don't just build AI; I build reliable AI. While the world focuses on prompt engineering, I focus on the reliability layer: grounding, hallucination mitigation, and adversarial robustness. I ship AI features that don't just look cool in a video, but actually move business metrics and earn user trust.
Currently obsessing over: Agentic Workflows, Local-First Intelligence, and AI Red-Teaming.
My first principle is building from scratch to understand the core—from MLOps to AIOps—and my end goal is creating high-impact open-source contributions that empower the global AI community.
| Area | My Approach |
|---|---|
| Reliable LLM Systems | RAG pipelines with strict grounding and automated eval harnesses. |
| Agentic Workflows | State-aware agents using LangGraph for complex, multi-step reasoning. |
| Local-First AI | Privacy-centric on-device models via Ollama (Zero latency, Zero API costs). |
| AI Security | Hardening LLMs against prompt injection and data leakage. |
| AI Products | Full-stack implementation from GPU kernels to Next.js frontends. |
Core Languages
AI / Machine Learning
Infrastructure & Web
🤖 Miii — Local AI Assistant · Open Source
The privacy-first AI agent. A fully on-device assistant with agentic workflows. No cloud, no latency, no API keys. Built for those who want GPT-level intelligence without sacrificing their data.
Ollama LangGraph Python Local-First
🐚 Miii CLI — The Local AI Shell · Open Source
Your terminal, supercharged. Bring LLMs directly into your shell. Automate boring tasks, query logs, and get instant intelligence without ever leaving the terminal.
CLI Local AI TypeScript Developer Experience
🛡️ mii-ai-security · Open Source
The armor for your LLMs. Production-ready security skill packs to prevent prompt injections and RAG leakage. Installable via npx miii-security add <skill>.
AI Security Adversarial Robustness LangChain LlamaIndex
🌐 AI Portfolio Assistant · Open Source
Interactive Experience. Why read a CV when you can chat with a digital twin? A real-time AI interface that knows my entire history and project depth.
Next.js OpenAI TypeScript Vercel
40,000+ users across products I've built
25+ countries reached
60% latency reduction in AI pipelines
35% improvement in user outcomes
"AI is only useful if you can trust it."
The gap between a demo and a product is Trust. I build the reliability layer—the evals, the guardrails, and the grounding—that turns a probabilistic model into a dependable tool.