Director · Product Infrastructure & AI 🏗️
I lead teams that build AI systems capable of operating at scale — not prototypes, not proofs of concept. Infrastructure that holds, agents that deliver, and models fine-tuned for the real world. 🌍
🔬 LLM Fine-tuning | 🤖 AI Agents | 🧠 Context Engineering | 🔗 MCP / A2A | ⚙️ AI Infrastructure | 📊 Data Engineering
Applied AI is what I do. I am hands-on by choice — because the gap between what a model can do and what it actually does in production is where most projects fail, and I would rather close that gap than manage around it. 🎯
I approach every model like a high-performance collaborator: give it the right context, the right tools, clear constraints, and get out of the way. My job is to build the systems that make that possible at scale. 🚀
I am invested in the developer community because great AI tooling should not be locked behind enterprise budgets. I share what I build, openly. 🤝
01 · 🔬 Fine-tuning open models QLoRA, LoRA adapters, and Mixture-of-Experts architectures for domain-specific production agents. All experiments are publicly available on HuggingFace under the same handle.
02 · 🤖 Production-grade AI agents Not novelties — agents that are measurably useful, defensible to a CFO, and built to operate without supervision. Scalable architecture is non-negotiable.
03 · 🔗 MCP and A2A protocols Unified context is the difference between an agent that coordinates and one that hallucinates in isolation. I build the connective tissue that makes multi-agent systems actually work.
04 · 🛠️ Infrastructure as code Secure, consistent, and reproducible infrastructure. I sleep soundly because the disaster recovery plan is tested, not assumed.
05 · 📊 Data engineering The discipline that started everything. I still model every system as an ER diagram — and I am not apologising for it.
- 💼 Current role: Director, Product Infrastructure & AI — SaaS, heavy asset industries
- 🎓 Education: BSc Software Engineering · MBA · PhD from the streets 😄
- 🔭 Outside work: 🍕 Pizza from scratch · 🌲 Wilderness · 🎬 Film · 💻 Vibe coding
- 🛠️ Open source: github.com/irfanalii · huggingface/irfanalii
🎬 Testing Gemini 2.5 Pro as an agentic video editor — what actually happened The timeline was full of hype. I built something instead. An honest account of what the model can and cannot do in a real agentic workflow.
👥 I staffed my side project with six AI agents named after real colleagues No contractors. No employees. Just me and a team of specialized agents, each designed to embody the working style of the best people I have worked with.
