MD. Nazmuddoha Ansary

Lead AI Engineer
Hiperdyne, Tokyo, Japan (Remote)
Sr. ML Engineer, Apsis Solutions Limited, Dhaka (Remote)
Ex-Machine Learning Engineer, Markopolo.ai, Dover, Delaware
Ex-Researcher (Deep Learning), Royal IT Limited, Dhaka
Ex-Research Assistant, LEGOS, CNES, Toulouse, France
Ex-R&D Software Engineer, CelltronEMS, Dhaka
__init__

I passed my B.Sc in EEE from BUET (Bangladesh University of Engineering Technology). Currently leading AI development on multimodal video and speech systems for Japanese and English, from training custom deep learning models to serving them in production with Triton, TensorRT and vLLM. My work spans OCR, speech, document AI and RAG systems — including a champion Bangla handwritten OCR and a patent-pending colour analysis algorithm. My interests also include image processing, geospatial analysis and computer security. From solving CTF's to playing with open source Quantum Computers, from looking into the secrets of Pascal's triangle to exploring the dark web, I find myself fascinated by Science, Technology, Engineering and Mathematics.

Email Github Kaggle

__experience__

Lead AI Engineer - Hiperdyne, Tokyo, Japan (Remote)

April 2023 - present
Leading video and speech processing development: native multimodal AI and deep learning models for Japanese and English
ONNX / TensorRT conversion and concurrent model serving with Python / C++ backends via Triton
Multimodal model deployment with vLLM: quantisation, benchmarking, fine-tuning and evaluation for production APIs
Patent application in Japan (custom colour analysis algorithm) | AI City Challenge with Nvidia Japan (12 hours to 3 minutes) | Safety-critical AI systems for a confidential client

---

Sr. ML Engineer - Apsis Solutions Limited, Dhaka, Bangladesh (Remote)

April 2021 - present
Leading all AI/ML/DL initiatives across projects with teams in Bangladesh, Malaysia, India and Singapore
Custom OCR (eKYC, Memo) and image analysis systems (field force verification, planogram/POSM detection)
Automation agents: chatbots, document analysis, dynamic Graph RAG systems
Champion, AI For Bangla 1.0 (Bangla handwritten OCR) | Custom systems for BATB, Akij, Medela, Abbott

---

Machine Learning Engineer - Markopolo.ai, Dover, Delaware

August 2020 - March 2021
Core NLP tools for customer feedback and social media analysis (sentiment, intent, NER, TF-IDF word clouds)
Facebook Graph API based data extraction | competitor analysis for local clients

---

Researcher (Deep Learning) - Royal IT Limited, Dhaka, Bangladesh

January 2018 - January 2021
Deep learning for environmental forecasting: flash flood water level forecasting (Meghna and Borak basins, BMD data)
Satellite image based Crop Information System (CIS)
Project proposals and scoping for SPARSO, Ministry of Agriculture and ICT Division

---

Research Assistant (RA) - LEGOS, CNES, Toulouse, France

July 2018 - October 2018
Sentinel-2 based Digital Elevation Model for coastal Bangladesh with a novel water body refinement algorithm
Featured by CNRS, Europe's largest fundamental science agency

---

R&D Software Engineer - CelltronEMS LTD, Dhaka, Bangladesh

May 2017 - December 2017
Application security: binary exploitation, reverse engineering, keygenning, parser differentials
.hta (CVE-2017-0199, CVE-2017-8570) and VBA-Script vulnerability assessment for MS-Office 2016

---

Experiment Lead - Bengali.ai, Dhaka, Bangladesh (Voluntary)

April 2021 - present
OCR R&D for 7 Indic languages with a grapheme-based tokeniser; Bangla spell/grammar checker, ASR, document layout analysis, Bangla RAG for historical documents
BengaliAI-speech: community competition winners selected by Google and Kaggle (sponsorship secured) | Co-organised DL Sprint (BUET), IUT Datathon | Mentor for national and international Kaggle competitions

---

SKILLS

Languages: Python, C/C++, Rust, MATLAB, C#
AI/ML: PyTorch, TensorFlow, Keras | MLOps & Serving: vLLM, Nvidia Triton, TensorRT, FastAPI, Azure DevOps (CI/CD)
Security: gdb, radare2, metasploit, owasp-zap, arpspoof | Geospatial: xarray, geopandas, rioxarray, QGIS
OS: Linux (Ubuntu, Kali), Windows | Spoken: English, Bangla

__research__

Journal

Improvement of system strength under high wind penetration: A techno-economic assessment using synchronous condenser and SVC.

Nahid-Al Masood, S. U. Mahmud, Nazmuddoha Ansary, S. R. Deeba.
Energy, Volume 246, 1 May 2022, 123426 [Elsevier].
Paper Link

---

High-resolution intertidal topography from sentinel-2 multi-spectral imagery: synergy between remote sensing and numerical modeling.

Md Jamal Uddin Khan , ​ Nazmuddoha Ansary , Fabien Durand , Laurent Testut , Marufa Ishaque, Stéphane Calmant , Yann Krien , A.K.M. SaifulIslam , Fabrice Papa .
MDPI [Remote Sensing] - Impact Factor- 4.118, Funding: CNES (French Govt. Space Agency) through the“BANDINO” TOSCA grant.
Paper Link Github

Conference

Unicode Normalization and Grapheme Parsing of Indic Languages.

Nazmuddoha Ansary, Quazi Adibur Rahman Adib, Tahsin Reasat, Asif Shahriyar Sushmit, Ahmed Imtiaz Humayun, Sazia Mehnaz, Kanij Fatema, Mohammad Mamun Or Rashid, Farig Sadeque.
LREC-COLING 2024
Paper Link Github

---

OOD-Speech: A Large Bengali Speech Recognition Dataset for Out-of-Distribution Benchmarking.

Fazle Rabbi Rakib, Sheikh Shafayat, Sadia Alam, ... , Nazmuddoha Ansary, ... , Farig Sadeque, Tahsin Reasat, Asif Sushmit, Ahmed Imtiaz Humayun.
INTERSPEECH 2023
Paper Link Dataset

---

BaDLAD: A Large Multi-Domain Bengali Document Layout Analysis Dataset.

Md. Istiak Hossain Shihab, Md. Rakibul Hasan, Mahfuzur Rahman Emon, Syed Mobassir Hossen, Md. Nazmuddoha Ansary, Intesur Ahmed, Fazle Rabbi Rakib, Shahriar Elahi Dhruvo, Souhardya Saha Dip, Akib Hasan Pavel, Marsia Haque Meghla, Md. Rezwanul Haque, Sayma Sultana Chowdhury, Farig Sadeque, Tahsin Reasat, Ahmed Imtiaz Humayun, Asif Shahriyar Sushmit.
ICDAR 2023
Paper Link Github

---

A Sweet Recipe for Consolidated Vulnerabilities: Attacking a Live Website by Harnessing a Killer Combination of Vulnerabilities for Greater Harm​.

Mazharul Islam, ​ MD. Nazmuddoha Ansary ​, ​Novia Nurain, Salahuddin Parvez Shams and A. B. M. Alim Al Islam.
NSysS-2018
Paper Link

---

Analysis of Modular Multilevel Converter Based Solid State Transformer Implementing Hybrid Control​.

Rajesh Paul,​ ​ Nazmuddoha Ansary​, Abdur Rokib, Irrtisum Khan and Shamim Reza
ICECE-2018
Paper Link Github

---

Intertidal digital elevation model (DEM) of CHITTAGONG using high-resolution SENTINEL 2 imagery and tidal model.

Nazmuddoha Ansary​ , Md. Jamal Uddin Khan, Fabien Durand, A.K.M. Saiful Islam.
ICWFM 2019, Funding: CNES, French Embassy, Dhaka, LEGOS, France
Github

---

R3Diver: Remote robotic rescue diver for rapid underwater search and rescueoperation.

Fattah,S.A,Abedin,F​,​Ansary,M.N ​ , Rokib,M.A,Saha,N,& Shahnaz,C.
IEEE TENCON 2016 Singapore
Paper Link

---

Particle Size Analysis by Image processing Technique.

Khaled M. Ehsan, Md. Raquibul Hossain, Tanvir Manzur,​ Ansary Shohag​ , Nazifa Tabassum.
BUET-ANWAR ISPAT 1st Bangladesh Civil Engineering SUMMIT 2016, At Bangladesh, Volume: SS215
Paper Link Github

__projects__

MLOPS AND LLM SERVING

Github Triton TensorRT-LLM Deployment for LLama 3.1 8B

---

Github Triton TensorRT Deployment for JinaV3 Embeddings

---

COMPUTER VISION AND OCR

Github Bangla-English Dual OCR

---

Github Bangla Handwritten OCR - Champion, AI For Bangla 1.0

---

Github Sentinel-2 Satellite Data Handling Library

---

Github TPU-keras BusterNet for Image forgery Detection

---

Github Wire Detection in Robotic Disassembly

---

NLP AND SPEECH

Github Grapheme Parser for Indic Languages

---

Github Bangla Unicode Normalization

---

Kaggle Romanized Bangla To Bangla Transliteration

---

Kaggle Wave2vec2 TPU-based Transfer Learning

---

SERIES DATA AND QUANTUM

Github Water level modeling for flood forecasting

---

Github Memory Efficient TFRecord for Small Molecule Drug Discovery

---

Github Quantum Image Representation With BitString Compression for Binary Images

---

Kaggle Hybrid Quantum Convolutional Networks

---

COMPUTER AND NETWORK SECURITY

Github "Hack Me" Exercise Solution - MUN

Binary exploitation, keygenning and reverse engineering walkthrough

---

Binary Hacking Training Presentation

Bypassing a password checking binary
Ways to know the actual password
REVERSE ENGINEERING, KEYGEN, PARSER DIFFERENTIAL, and more

---

.hta (​ CVE-2017-0199 and CVE-2017-8570​ ) and VBA-Script vulnerability binary assessment for evading virus definitions on MS-WORD and MS-POWERPOINT exploiting OLE (Object Linking and Embedding) vulnerabilities in MS-OFFICE 2016.

Valid Till-2017 December

---

Improvised Technique with SSLSTRIP, dns2proxy, and ARP spoofing for getting HTTPS data in the local Wi-Fi-network​

Valid Till-2017 October

---