Open to Opportunities

Hi, I'm Niloy
Sarker

CS Undergrad · AI/ML Researcher · Backend Engineer

Building intelligent systems at the intersection of AI research and software engineering. Passionate about RAG pipelines, NLP, and transforming academic ideas into production-ready solutions.

View My Work Get In Touch

3.51

CGPA at SUST

1st

IEEE Paper Author

Years Leading RoboSUST

Focus

AI/ML Research

IEEE

COMPAS 2025

Rank

1st @ SUST in Computer Vision DL Enigma

About Me

Turning Ideas Into
Intelligent Systems

Years
RoboSUST

CS Undergrad with a Passion for AI Research & Engineering

I'm a Computer Science and Engineering student at Shahjalal University of Science and Technology (SUST), with deep interests in AI/ML, NLP, RAG systems, and backend engineering. I enjoy building systems that bridge the gap between cutting-edge research and practical, deployable solutions.

Beyond academics, I've led RoboSUST as President for 3+ years, driving R&D initiatives, organizing technical events, and mentoring teams in robotics and IoT competitions. This leadership experience has shaped my ability to coordinate, communicate, and deliver under pressure.

University SUST (CSE)

CGPA 3.51

Leadership President, RoboSUST

Publication IEEE COMPAS 2025

Download Resume

Skills & Expertise

My Technical Arsenal

From building RAG pipelines to competitive ML — here's what I bring to the table.

AI / ML & Deep Learning

End-to-end ML pipelines, object detection, ASR, OCR, VLMs, and transformer-based architectures.

PyTorch XGBoost RT-DETR Transformers Fine-Tuning

RAG & LLM Systems

Production-ready RAG pipelines with semantic chunking, vector stores, and citation-grounded responses.

LangChain Qdrant Ollama Sentence-Transformers RAG

Backend Engineering

Scalable REST APIs, database design, ORMs, and production-grade backend architectures.

FastAPI PostgreSQL SQLAlchemy Python REST APIs

CS Fundamentals

Strong foundation in DSA, OOP, problem solving, and competitive programming patterns.

DSA OOP Problem Solving C++ Python

IoT & Robotics

Hands-on experience with IoT systems, embedded hardware, and robotics through 3 years at RoboSUST.

IoT Arduino Sensors Embedded Systems

Leadership & Coordination

3+ years leading RoboSUST as President and R&D Secretary — managing teams, events, and research.

Team Lead Event Mgmt Mentoring R&D Strategy

Featured Projects

Things I've Built

From RAG-powered research assistants to hackathon-winning platforms.

RAG System

PaperSense — Research Paper Assistant

Production-ready RAG system for academic papers with section-aware PDF parsing, semantic chunking, Qdrant-based retrieval, and citation-grounded LLM responses via scalable APIs.

FastAPI Qdrant Sentence-Transformers Ollama PostgreSQL

Code Demo

Backend

HackMatch — Hackathon Platform

Centralized hackathon platform for Bangladeshi universities where companies scout and recruit talent, featuring smart discovery, team workflows, and a scalable backend.

FastAPI Python SQLAlchemy PostgreSQL

Code Details

Full Stack

Smart Waste Management — Code Samurai 2024

End-to-end waste management system for Dhaka with React-based citizen features for reporting, tracking, and engagement. Built during Bangladesh's largest hackathon.

React MongoDB JavaScript CSS

Code Details

Machine Learning

ML & Deep Learning Experience

Competition-tested ML pipelines with real-world accuracy gains.

Object Detection

DL ENIGMA — Real-World Object Detection

Top performer (1st among SUST) by optimizing real-world data pre-processing, adopting RT-DETR for deployment readiness, and applying Weighted Box Fusion (WBF) ensembling — yielding a 4% accuracy boost. Placed 4th overall out of 46 teams at SUST CSE Carnival.

+4% Accuracy Boost · Ranked 1st @ SUST

Multimodal AI

OCR–VLM Multimodal Analysis Pipeline

Developed an OCR–VLM multimodal analysis pipeline for the CUET PoliMeme Kaggle Datathon. Experimented with multiple OCRs before adopting NanoNet (achieving 78% accuracy) and fine-tuned a Vision Language Model with self-curated processed data and Chain-of-Thought prompting.

78% OCR Accuracy · CoT Prompting

Speech & NLP

ASR–LLM Pipeline for Regional Speech

Developed an ASR-LLM pipeline for the SHOBDOTORI CUET Datathon. Evaluated multiple STT models, finalized Tugstugi, and fine-tuned an LLM for regional-to-standard Bangla text conversion — achieving 85% accuracy.

85% Accuracy

Fraud Detection

End-to-End Fraud Detection Pipeline

Built a complete fraud detection system with extensive feature engineering, robust data wrangling, and a GPU-accelerated XGBoost ensemble — achieving 90.4% accuracy on real-world transaction data.

90.4% Accuracy · GPU XGBoost

Research

Academic Research

Peer-reviewed publications and ongoing thesis work.

Accepted

Bangla Speech to IPA Transcription: A Multi-Stage Transformer-Based Approach

2nd IEEE Conference on Computing Applications and Systems (COMPAS 2025)

First-author paper proposing a multi-stage transformer pipeline for converting Bangla speech into International Phonetic Alphabet (IPA) transcriptions. The system leverages cascaded transformer modules for robust phonetic mapping from audio input.

Ongoing Thesis

Transformer-Based Sign Language Recognition with Gloss Sequence Generation

B.Sc. Thesis Research — SUST CSE

Investigating transformer architectures for continuous sign language recognition from video, with a focus on generating accurate gloss sequences. Combines computer vision with sequence-to-sequence modeling for accessibility applications.

Ongoing Thesis

Bangla Tabular OCR for Structured Table Reconstruction

B.Sc. Thesis Research — SUST CSE

Developing an OCR system specifically tailored for Bangla text in tabular formats, enabling accurate extraction and structured reconstruction of table data from scanned documents and images.

Achievements

Recognition & Awards

Highlights from competitions, hackathons, and academic events.

Ranked 1st Among SUST

DL ENIGMA — SUST CSE Carnival

Top performing SUST student in Object Detection competition. Placed 4th out of 46 teams overall.

Finalist — Top 10%

CODE SAMURAI — Bangladesh's Largest Hackathon

Selected as finalist from 400+ competing teams in Bangladesh's premier hackathon competition.

Finalist — Top 15%

UIU Bhashamul

Demonstrated skills in IPA-to-text conversion, data pre-processing, and sequence modeling.

IEEE Paper Accepted

COMPAS 2025 — IEEE Conference

First-author paper accepted at the 2nd IEEE Conference on Computing Applications and Systems.

President — RoboSUST

3+ Years of Leadership

Led RoboSUST as President and former R&D Secretary, driving robotics & IoT research initiatives.

Multiple Datathon Finalist

CUET PoliMeme · SHOBDOTORI · Others

Consistently placed in top tiers across multiple competitive ML datathons and challenges.

Get In Touch

Let's Work Together

Open for research collaborations, freelance projects, and full-time opportunities.

Reach out — I'd love to hear from you.

Whether you have a project idea, research collaboration opportunity, or just want to connect — feel free to reach out. I typically respond within 24 hours.

niloysarker0033@gmail.com

github.com/niloysarker

Hi, I'm Niloy Sarker

Turning Ideas IntoIntelligent Systems