Photo of Emtiaz Uddin Ahmed

Emtiaz Uddin Ahmed

Senior Machine Learning Engineer

Tulip Tech · Dhaka, Bangladesh

  Download CV

Biography

Hi! Welcome to my website :)

I'm a Senior Machine Learning Engineer at Tulip Tech in Dhaka, Bangladesh, where I build agentic and voice-AI systems for healthcare. I received my B.Sc. in Computer Science and Engineering from Rajshahi University of Engineering & Technology (RUET).

My research asks how modern AI systems fail under real-world conditions, and how to detect and repair those failures without retraining: when do LLMs and agents state things they were never told, how much do vision–language models know about their own errors, and why do multilingual speech models stumble on rare letters and scripts? These questions come straight from the systems I have built in industry, from real-time Bangla voice agents to multi-agent assistants for healthcare.

I am applying to PhD programs starting Fall 2027. If our interests overlap, I would love to hear from you.

Interests

  • Reliable LLMs & agents
  • Uncertainty & interpretability in VLMs
  • Multilingual & low-resource speech
  • Evaluation methodology

Education

  • B.Sc. in Computer Science and Engineering, 2022Rajshahi University of Engineering & Technology

Updates

Recent news

Research

What I work on

  Reliable LLMs and agents

When do language models and agents state things they were never told? I study fabrication and abstention in LLM outputs, and how agents that keep long-term memory handle the difference between what a user said and what the model inferred. I look for lightweight fixes that hold up across many models.

hallucinationabstentionagent memory

  Uncertainty and interpretability in vision–language models

How much does a vision–language model know about its own errors? I study calibration and training-free hallucination detection, and whether mechanistic diagnoses of VLM failures actually turn into fixes that beat simple baselines.

calibrationhallucination detectioninterpretability

  Multilingual and low-resource speech

Text-to-speech for languages and scripts that large models underserve, starting from Bangla, with a focus on careful evaluation of where multilingual models break.

TTSBanglaevaluation

  Papers from these projects are under double-blind review, so they are not listed here yet. Drafts are available on request.

Experience

Research and industry

Hugging Face

Open models, datasets and live demos for Bangla speech

EMTIAZZ

Main account · models and live demos

EmtiazUdd

Research data and checkpoints

Honors and Awards

Competitive programming and hackathons

IUT 11th National ICT Fest IUPC

Ranked 18th of 150+ teams

2024

HackNSU

Top-10 finalist of 100+ teams; real-time disaster-relief platform

2023

ICPC Asia Dhaka Regional

Ranked 33rd of 150+ teams

2022

ICPC Asia Dhaka Regional Preliminary

Ranked 46th of 1,600+ teams

2022

National Hackathon on Frontier Technologies

Finalist of 2,000+ teams; AI traffic monitoring

2022

RUET CodeSmash

Ranked 5th of 100 contestants

2019

3,000+ problems solved · LeetCode (max 1986) · Codeforces (Specialist, max 1525) · CodeChef (max 1731) · AtCoder

Skills

Tools I use

Research

Experimental designStatistical testingProbingUncertainty quantificationSpeech evaluation

Machine learning

PyTorchHugging FacePEFT / LoRAJAXvLLM

Systems

LLM agentsRAGLangGraphLiveKitTTS / STTFastAPI

Infrastructure

PythonC++DockerKubernetesAWSKafka

Contact

Get in touch

I'm always happy to talk about research, PhD opportunities, or collaborations.