Niloy Kumar Kundu
A person of limitless vision (Niloy), carrying the youthful spirit of growth and responsibility (Kumar), and rooted in a proud Bengali heritage (Kundu).
AI Engineer & Researcher
Bridging research and real-world impact in AI.
My curiosity lies at the intersection of AI for Software Engineering, Software Security, Large Reasoning Models for Source Code, and the combination of Reinforcement Learning with reasoning-driven ML systems. I am also drawn to Medical Image Processing, AI for Health and the use of LLMs / VLLMs. Recently, I have been working on GenAI applications and experimenting with agentic AI solutions that bring together reasoning, retrieval, and tool-use. Beyond prototyping, I enjoy making these ideas practical by deploying ML systems in the cloud and exploring how Azure AI services can support research-to-application transitions. What excites me most is building AI systems that do more than perform well, they can reason, adapt, and support human decision-making in meaningful ways.
I completed my B.Sc. in Computer Science and Engineering (Major: Data Science) at United International University, graduating with a CGPA of 3.87/4.00 and ranking in the top 4% of my cohort. My thesis, Enhanced Speech Emotion Recognition with Efficient Channel Attention-Guided Deep CNN-BiLSTM Framework, focused on building a lightweight model that achieved strong results across multilingual datasets. I also served as a Research Assistant on a funded project (IAR Grant: BDT 500,000) in collaboration with Brandenburg University, Germany, where I worked on NSCLC subtype classification using radiomics features. My research contributions include publications on precision agriculture (Computers and Electronics in Agriculture, IF: 8.9), medical image analysis (ICCIT 2023), and federated learning (IEEE I2CT 2023). Along the way, I received multiple merit-based scholarships, served as a Teaching Assistant for Data Structures and Algorithms, and led initiatives as Instructor and Head of Research Assistants at the UIU App Forum. I also enjoyed competitive programming, placing 5th in the UIU Coders’ Combat 3.0: Intra-University Programming Contest.
Looking ahead, I am passionate about advancing safe and reliable AGI research. In the near term, I aim to contribute to reasoning-augmented LLMs for secure software engineering, reproducible ML benchmarks, and interpretable medical AI. In the long run, my vision is to establish a research group that not only develops explainable and security-aware AGI systems but also fosters collaboration, mentorship, and open science. My goal is to help build AI that strengthens both the tools we use to engineer software and the systems we rely on in healthcare, making them more transparent, trustworthy, and impactful.
Looking for MS/PhD opportunities. Open to talk about both academic and industrial collaborations and opportunities.
Experiences
Professional Experiences
For details visit here.
August 2025 - Present (Remote, Full-Time)
Currently engaged in cutting-edge AI projects including a Multi-Agent Conversational Product Recommender System (Chatbot), advanced AI-powered product recommenders, and an intelligent image search engine for multiple client solutions.
- Deployed an Agentic AI–based Multi-Agent Conversational Product Recommender Chatbot on Microsoft Azure, ensuring scalability, security, and high availability in production.
- Integrated Azure AI Vision and Document Intelligence services within Azure AI Studio for OCR and visual analysis.
- Leveraged Azure AI Search to enable advanced retrieval-augmented generation (RAG) and semantic search capabilities.
- Integrated diverse AI-powered tools to streamline data extraction, processing, and knowledge integration pipelines.
- Designed dynamic workflows with LangGraph to intelligently manage user queries and perform real-time interactions with a MySQL database.
- Evaluated and integrated ChatGPT using the OpenAI API key to enhance response accuracy and conversational quality.
- Developed a comprehensive evaluation pipeline to benchmark CRS performance using standardized metrics and user-centric KPIs.
- Spearheaded R&D on an AI-driven Conversational Recommender System (CRS) enabling natural language–based product search and customization.
May 2025 - August 2025 (Hybrid, Full-Time)
Worked on multiple in-house AI based products. Brainstormed on in-house product development and also for the clients.
- Developed and deployed end-to-end Machine Learning (ML) pipelines, including data preprocessing, model training, and production deployment.
- Worked on Large Language Models (LLMs) and Explainable AI (XAI) to deliver transparent and interpretable business solutions.
- Built and deployed multiple Azure-based ML services for clients, leveraging Azure Machine Learning and API endpoints for scalable solutions.
- Designed and implemented Agentic AI solutions using LangGraph to enable autonomous decision-making and intelligent workflow automation.
- Collaborated with cross-functional teams to brainstorm, design, and prototype AI products for both in-house initiatives and external clients.
- Implemented MLOps best practices including CI/CD pipelines, monitoring, and logging systems to ensure production reliability and maintainability.
December 2023 - April 2025 (Onsite, Full-Time)
Led applied research and development of machine learning and IoT solutions for precision-agriculture projects in collaboration with the Bangladesh Agricultural Research Council (BARC) and the Institute for Advanced Research (IAR).
- Designed and deployed ML pipelines for IoT-driven smart irrigation systems based on the Alternate Wetting and Drying (AWD) technique to improve rice cultivation sustainability.
- Developed computer vision models (custom CNNs) to estimate water height from AWD pipe images; managed data collection, labeling, augmentation, and cleaning using field IoT feeds.
- Built real-time predictive models for automated water-level monitoring and decision-making in agricultural fields.
- Created an AI-powered Agri-Advisory Chatbot using a RAG architecture with Pinecone vector database for fast and contextually relevant knowledge retrieval.
- Integrated speech-to-text (ASR) and text-to-speech (TTS) modules to enable voice-based interactions, improving accessibility and non-text engagement by ~60%.
- Ensured production readiness through model validation, CI/CD packaging, and monitoring/logging for reliability and maintainability.
- Published peer-reviewed research on water-height estimation methodology, demonstrating scientific rigor and practical impact for AWD management.
- Collaborated cross-functionally with agronomists, IoT engineers, and stakeholders (BARC, IAR) to convert research outcomes into usable, scalable tools—driving measurable improvements in irrigation decision support and farmer usability.
Undergraduate Research Experience
July 2023 - January 2024
Contributed to an international research project on lung cancer diagnosis, applying machine learning to medical imaging. Helped build a system that analyzes CT scans to distinguish between cancer subtypes with high accuracy, supporting more effective and data-driven healthcare solutions.
- Conducted radiomics-based analysis of CT scan datasets from 150 lung cancer patients to classify NSCLC subtypes (Adenocarcinoma vs. Squamous Cell Carcinoma).
- Extracted over 100 radiomic features from segmented tumor regions using 3D Slicer, excluding shape-based features to focus on texture and intensity.
- Applied advanced feature selection techniques such as LASSO and Recursive Feature Elimination (RFE) to refine datasets.
- Implemented and compared multiple machine learning algorithms (Random Forest, Decision Tree, Logistic Regression) for subtype classification.
- Achieved 92% classification accuracy and 94% AUC with the Random Forest model, demonstrating strong predictive performance.
- Collaborated with cross-institutional teams (UIU, Brandenburg University, and South Asia Centre for Medical Physics & Cancer Research) to align research outcomes with clinical relevance.
- Worked under the supervision of Dr. Hasin Anupama Azhari
September 2022 - April 2023
Developed a lightweight dual-branch CNN-BiLSTM model with attention mechanisms for multilingual speech emotion recognition, achieving high accuracy across five benchmark datasets.
- Engineered and pre-processed multilingual audio datasets using Librosa, extracting features such as MFCCs, Mel-spectrograms, ZCR, and RMS.
- Designed a dual-path CNN-BiLSTM architecture enhanced with ECA-Net attention to capture both local spectral patterns and long-term temporal dependencies in speech.
- Applied data augmentation, hyperparameter tuning, and cross-validation to maximize generalization and robustness.
- Benchmarked model performance on five public SER datasets, achieving superior results compared to baseline approaches.
- Worked under the supervisions of Md Rayhan Ahmed
Volunteer Experiences
August 2022 – September 2023
Volunteered as Head of R&D during my final year, where I guided innovation-driven projects and mentored peers.
- Led the Research & Development wing, overseeing multiple student-driven technical and research projects.
- Directed multiple research-based student projects, ensuring structured planning, execution, and delivery.
- Mentored junior members in research methodology, problem-solving, and software development practices.
- Organized technical discussions and knowledge-sharing sessions to promote a culture of continuous learning.
- Collaborated with faculty members to integrate research findings into student projects and workshops.
- Strengthened the club’s culture of professionalism, teamwork, and technical excellence through leadership and collaboration.
March 2020 – September 2023
Volunteered as an instructor and mentor, training junior students in programming and project development.
- Conducted the “Programming for Beginners” workshop 3 times, teaching 150+ students from basics to advanced C concepts.
- Designed the full course curriculum and delivered 14 structured sessions per batch.
- Guided participants in building CLI-based projects and led project showcases to highlight their work.
- Provided one-on-one mentorship to students struggling with core programming concepts.
- Fostered a collaborative learning environment that boosted students’ confidence in problem-solving and coding.
- Inspired students to pursue competitive programming and practical applications of C.
Publications
For details visit Google Scholar.Published
Accurate water level monitoring in Alternate Wetting and Drying rice cultivation using attention-based ConvNeXt architecture
Ahmed Rafi Hasan, Niloy Kumar Kundu, Saad Hasan, Mohammad Rashedul Hoque, Swakkhar Shatabda
Published in - Computers and Electronics in Agriculture (Volume 234,Pages 110216) [Impact Factor: 8.9]
Attention Based Feature Fusion Network for Monkeypox Skin Leison Detection
Niloy Kumar Kundu, Mainul Karim, Sarah Kobir, Dewan Md Farid
Published in - 2023 26th International Conference on Computer and Information Technology (ICCIT)
Implementing Federated Learning Based on Rainforest Model
Mainul Karim, Niloy Kumar Kundu, Dipu Saha, Sarah Kabir, Sumaiya Akter Mim, Dewan Md Farid
Published in - 2023 IEEE 8th International Conference for Convergence in Technology (I2CT)
Projects
Personal Projects
Developed Multi-Model LLM Question Answering App, a Dockerized platform that allows users to select from multiple language models and engage in context-aware, memory-enabled conversations. The system combines a Streamlit interface with an Ollama LLM server, while an Nginx reverse proxy efficiently routes requests. Orchestrated using Docker Compose, the deployment is scalable, portable, and GPU-ready, providing a robust environment for LLM experimentation and production-ready applications.
Built an interactive Streamlit-based application that enables users to chat with their CSV and Excel files using natural language queries. The system integrates PandasAI with Groq LLMs for intelligent data exploration and visualization, allowing users to generate insights, tables, and plots instantly. Designed a custom response parser to seamlessly render dataframes and charts within the app, ensuring a smooth user experience. The solution supports multiple file uploads, real-time analysis, and has been containerized and deployed using Docker for portability and scalability.
Certifications
MOOCs
Claude Code: A Highly Agentic Coding Assistant!
Deeplearning.ai
View Certificate
AI Agents Fundamental
Hugging Face
View Certificate
Supervised Machine Learning: Regression and Classification
Coursera
View Certificate
Intro to Machine Learning
Kaggle
View Certificate
Python
Kaggle
View Certificate
