Quazi

Quazi Marufur Rahman


Available for consulting. I help teams turn LLM and agent prototypes into systems that hold up in production. If your AI works in a notebook but breaks with real users, that’s the gap I close. I take a small number of engagements at a time.  Get in touch.

I’m Quazi Marufur Rahman — an AI engineer with a PhD in robotic vision and 10+ years shipping machine learning into production. I specialize in LLM and agentic AI systems: agents, RAG, and fine-tuning that work outside the demo.

Right now I’m a Technical Specialist (Computer Vision & AI) at Transoft Solutions, leading R&D on multi-agent LLM systems and production RAG. Before that I was AI Team Lead at Advanced Mobility Analytics Group, delivering road-safety video analytics on AWS at real-world scale. My PhD work was on detecting when vision systems fail at run-time — which is still how I think about every system I ship: AI you can trust in production, not just in a slide.

I also run neuralwork, my independent AI product studio, where I build small, useful AI tools — and write up the hard parts on the Notepad.

Selected Results

  • 692+ crashes and 113+ injuries prevented — three production video-analytics platforms I led for proactive road safety.
  • 90%+ mAP in production — detection and tracking systems running under real-world constraints, not benchmark conditions.
  • 70% less manual reporting — agentic AI pipelines for automated engineering-data analysis and document synthesis.
  • 1,000+ hours of video/day — AWS processing infrastructure (Batch, SageMaker, Lambda, Step Functions) I architected.

What I Can Help With

  • LLM & Agentic AI Systems (primary focus) — agents, multi-agent orchestration, RAG pipelines, and tool-using LLM workflows that hold up in production, with evaluation and monitoring baked in.
  • LLM Fine-Tuning & Evaluation — domain adaptation, fine-tuning, and the eval harnesses that tell you whether it actually got better.
  • Production ML & Computer Vision — the supporting depth: detection, segmentation, tracking, MLOps, and end-to-end deployment on AWS.

How I Work

Most engagements fall into one of three shapes:

  • Advisory — review your roadmap, architecture, or a stalled system, and tell you where the real risk is.
  • Prototype — build a working, testable proof of concept to de-risk an idea fast.
  • Build — design and ship a production system end-to-end, with evaluation and monitoring.

I usually start with a short, fixed-scope first step so we both know it’s a fit before committing further. I work remotely and take a limited number of engagements alongside my full-time role.

Let’s Talk

Have an AI problem worth solving? Email me a few lines on what you’re building and where it’s stuck. If I can help, we’ll book a free 20-minute call to scope it — no pitch, no obligation.

📧  hi@qmaruf.me  ·  💼  LinkedIn