Skip to content
Get in touch

Lecturer at BRAC University · open to research collaboration

Nazmus Sakib

AI/ML Engineer & Lecturer

I build production LLM systems, including fine-tuning, RL post-training and retrieval pipelines, and teach the fundamentals behind them. Computer Science & Engineering graduate of BUET, with published research on memory architectures for conversational web agents.

Résumé
Portrait of Nazmus Sakib

About

Engineer, researcher, teacher

I work at the point where language models stop being demos and start being systems people depend on, and I teach the fundamentals that make that possible.

LLM post-training, end to end

I build the full pipeline: teacher-distilled dataset synthesis, supervised fine-tuning, then GRPO with composite reward design. I also report what actually happened. My GeoQL work is published as a negative result because the RL stage did not beat SFT, and that is worth documenting.

Retrieval systems that survive real data

Agentic and graph RAG in production for banking, legal, civil engineering and e-commerce domains: hybrid indexing, multimodal embeddings, reranking, and human-in-the-loop escalation when confidence drops.

Full-stack by default

A model is not a product. I ship the FastAPI and Django services around it, the SvelteKit and Next.js interfaces on top, the Flutter client beside it, and the Docker, Nginx and CI/CD that keep it running.

Teaching the fundamentals

As a Lecturer at BRAC University I teach the theory underneath the tooling. Explaining a concept to a room of students is the fastest way to find the gaps in your own understanding of it.

Competitive problem solving

Six national hackathon and contest wins, and an active competitive programming habit. Decomposing a messy problem under time pressure is a transferable skill, and it is the one I lean on most.