Zia Hasan

Machine Learning Scientist • Conversational AI • Generative Models

Self-Hosting Enthusiast & Tech Explorer • Lifelong Learner & Traveler

Welcome to my corner of the web. I run my own infrastructure using Proxmox, TrueNAS, and Docker.

10+ years building AI systems NLP • LLMs • RAG • CV Leadership & product delivery
GitHub LinkedIn Google Scholar

About

I build practical machine learning systems - especially conversational AI - bridging research and product. My work spans large language models, retrieval-augmented generation (RAG), agentic workflows, evaluation tooling, and applied ML for search and personalization.

I enjoy designing end-to-end pipelines (data → modeling → evaluation → deployment) and helping teams deliver robust, measurable improvements in real user experiences.

Focus Areas

  • • Multi-turn assistants & tool use
  • • Retrieval & ranking for enterprise search
  • • LLM fine-tuning and safety-minded evaluation
  • • Active learning & data-centric ML
  • • Multilingual NLP and NER / slot-filling

Highlights

Conversational AI at scale

Led core initiatives for multi-turn NLU and assistant experiences, including RAG and agentic routing patterns.

Evaluation & quality systems

Built automated evaluation frameworks and dataset curation workflows for iterative improvement.

Applied ML for search

Delivered production models for query understanding and correction, improving retrieval quality across markets.

Experience

Staff / Lead Machine Learning Scientist

Enterprise Platform

2020 – Present

  • Owned ML direction for conversational AI and NLU capabilities used in production assistants.
  • Designed multi-turn conversational search using RAG and agentic patterns (planning + routing).
  • Built automated evaluation for multi-turn dialogues and a pipeline for dataset curation.
  • Developed active-learning workflows to reduce labeling cost while improving classification quality.

Applied Scientist / ML Engineer

Consumer Search & Discovery

2019 – 2020

  • Developed query understanding and correction models using modern neural architectures.
  • Built large-scale behavior datasets to improve robustness and reduce noisy corrections.
  • Explored summarization techniques for low-resource settings.

Research Scientist

AI Research Lab

2014 – 2019

  • Worked on NLP and CV research, including paraphrasing, grammar correction, and segmentation.
  • Built applied ML for anomaly detection and predictive analytics in large-scale systems.
  • Partnered with engineering teams to bring prototypes into field deployments.
*Company names and select details are intentionally generalized for this public-facing page.

Selected Work

Agentic RAG & routing

Architected patterns for query planning, tool selection, and multi-agent routing for complex conversations.

LLM evaluation micro-judges

Designed evaluation components that grade multi-turn assistant behavior and enable iterative prompt/model improvements.

NER & enterprise entity normalization

Built entity extraction and normalization components tailored for domain-specific enterprise workflows.

Anomaly detection for time-series

Developed detection pipelines for operational signals, emphasizing precision/recall and deployability.

Skills & Technologies

Core

LLMs NLP Conversational AI RAG Active Learning Computer Vision Reinforcement Learning

Stack

Python PyTorch TensorFlow scikit-learn Java PySpark AWS / Azure CUDA (basics)

Education

Graduate Certificate (AI)

Stanford University

Advanced coursework in NLP and Reinforcement Learning

Ph.D. & M.A.Sc. (Electrical & Computer Engineering)

University of British Columbia

Optimization, system modeling, signal processing

B.Tech. (Electrical Engineering)

IIT Kanpur

Publications & Patents

Author of peer-reviewed work across machine learning and communications, with a strong citation record. Contributed to multiple patents (granted and pending) in areas including data labeling, prompt refinement, and AI workflow orchestration.

Details available upon request.

Get in Touch

I'm always open to discussing ML, AI, infrastructure, or interesting projects. The best way to reach me is through LinkedIn.