About
Hi, this is Shaun. I'm a machine learning engineer who builds and ships AI systems, from language models to the agents and pipelines that put them to work. At my day job I build agentic systems: LLMs that call tools, talk to databases, and hold up under real traffic, while also keeping them cheap to run. In my free time, I work on Bangla NLP to help close the low-resource gap, and I'm gearing up for postgraduate study to go further on both.
- Current role
- AI Engineer @ Synesis IT PLC
- Location
- Dhaka, Bangladesh
- Focus
- LLM · agentic AI · retrieval systems
- Publication
- BanSuite (EACL 2026)
- Looking for
- AI/ML roles, research collaborations, master's programmes
Experience
Jun 2026 – Present
Synesis IT PLC · Dhaka, Bangladesh
AI Engineer
Research and implement optimisation techniques for small and
large language models, including quantisation, to improve inference speed, reduce model
size, and increase deployment efficiency. Design and integrate AI-driven features into
Convay, an online meeting platform, using LLMs to deliver intelligent meeting
experiences and productivity enhancements.
Sep 2024 – Jun 2026
Giga Tech Limited · Dhaka, Bangladesh
Machine Learning Engineer
Designed an AI-powered web crawler to extract structured and
unstructured data from heterogeneous websites using adaptive parsing and LLM-based
content understanding. Developed an agentic RAG-based HR chatbot that automated
employee query resolution across internal systems, including SQL databases and vector
stores, using open-source LLMs. Trained and fine-tuned Bangla language models, including
BERT, GPT, T5, and LLaMA, on large-scale in-house corpora, improving linguistic
coverage, domain adaptation, and downstream performance on NER, POS tagging, and
question answering.
Jun 2024 – Aug 2024
Syntax Solution Limited · Dhaka, Bangladesh
Trainee Software Engineer
Helped build a recommendation engine, replacing keyword search
with a hybrid architecture that integrates semantic similarity with keyword-based
techniques. Built a retrieval-augmented generation chatbot pipeline to automate
repetitive customer queries.
Education
Undergraduate
World University of Bangladesh · Dhaka
B.Sc. in Computer Science & Engineering
Algorithms, data structures, databases, and software design — the
engineering foundation that everything since has been built on.
Currently preparing for postgraduate study
Skills
- Languages
- Python Rust SQL Bash
- Modelling
- PyTorch TensorFlow JAX Transformers Hugging Face NumPy
- Serving
- vLLM llama.cpp Triton ONNX
- Backend
- FastAPI
- Data
- Pandas Polars
- Database
- PostgreSQL MongoDB VectorDB
- MLOps
- Docker CI/CD
Research
EACL 2026 · System Demonstrations
A Bangla NLP platform bringing POS tagging, NER, shallow parsing, and
dependency parsing under one roof, trained on a large manually-annotated treebank —
around 90% F1/UAS, ahead of existing Bangla systems and general multilingual LLMs.
With Faisal Ahamed Khan, Jannatul Ferdous Tuli, Nabeel Mohammed, Mohammad Ruhul Amin,
Mohammad Mamun Or Rashid.
- Multi-agent systems — coordination, delegation, and control across many agents
- AI agents — tool use, planning, memory, and reliability under real workloads
- Large language models — training, adaptation, and efficient inference
- AGI — generalisation and reasoning beyond narrow, task-specific systems
Projects
Wires plain Python functions — and the agents inside them — into a
graph that runs branches in parallel, routes on conditions, and threads one typed
state object through the whole thing.
An async crawling library with sitemap discovery, browser-based
extraction, and LLM-powered structured output — all driven by a single Settings object.
The service that turns the library into a platform: a REST API for
auth, crawl jobs, settings and extracted data, with crawling handed to an async job
queue.
A PyTorch-backed package for rapid language-model development:
tokenisation, distributed training, and evaluation without rebuilding the same
scaffolding.
A dual-architecture pipeline for training and evaluating Word Sense
Disambiguation models, shipped with pretrained weights so you can start from a working
baseline.
Bangladesh's national Bangla NLP platform — corpus development,
syntactic annotation, low-resource language models, and downstream tasks built on top
of them.
Overleaf's editing experience, running entirely on your own
machine — a live LaTeX editor, a real compiler, and a real terminal in one window, with
no cloud, no account, and no multi-gigabyte TeX Live install.
Certifications
Generative AI with Large Language Models
DeepLearning.AI TensorFlow Developer
Transformer Models and BERT Model
Machine Learning Specialization
Mathematics for Machine Learning and Data Science
Data Visualization with Python
Master Git and GitHub: Beginner to Expert
Finding Your Professional Voice: Confidence & Impact
Contact
Open to AI/ML engineering roles, research collaborations, and master's programmes. Send me a message, I'll get back to you.