Tamal Hansda — AI/ML Engineer

01

About

I'm an AI/ML engineer working across generative AI, large language models, and computer vision. My approach is hands-on and foundational: I build core machine-learning machinery — automatic differentiation, optimizers, diffusion and vision pipelines — myself, so I understand exactly how and why they work. I hold an M.Tech in Vision & Intelligent Systems from IIT Kharagpur, and alongside my work I tinker with embedded ML and run a small home GPU lab for experimentation.

02

Experience

AI/ML Engineer

Benzy Infotech Pvt. Ltd.

Full-time · Jun 2025 – Present · Kochi, Kerala

Building and deploying machine-learning systems across generative AI, LLMs, and computer vision.

Research Intern

Indian Institute of Technology, Kharagpur

May 2024 – Jul 2024 · Kharagpur, West Bengal

Generated a ~4,500-image synthetic dataset of sketches and paired prompts with Stable Diffusion 2; fine-tuned the U-Net of a Latent Diffusion model using LoRA for parameter efficiency, and cut the FID score by 50% by further fine-tuning the VAE decoder with a perceptual loss.

Programmer Analyst Trainee

Cognizant

Full-time · 2023 · Pune, Maharashtra

Designed and built scalable ETL data pipelines with Informatica PowerCenter, IICS, and SQL.

Data Analyst

Cognizant

Internship · Mar 2022 – Aug 2022 · Remote

Worked with Google Analytics, Informatica PowerCenter, IICS, SQL, and PL/SQL.

Software Development Intern

Healthy Hearing Pvt. Ltd.

Internship · Sep 2020 – Dec 2020 · Remote

Back-end development and integration; migrated the website into a desktop application using the Electron framework.

Product Development Intern

Dronaz Artificial Intelligence & Robotics Pvt. Ltd.

Internship · Jun 2020 – Sep 2020 · Remote

Worked on LiDAR-based mapping and localization for a mobile robot — SLAM, Monte-Carlo Localization, and Python.

03

Education

M.Tech — Vision & Intelligent Systems

Indian Institute of Technology (IIT) Kharagpur

Jul 2023 – Jun 2025 · CGPA 8.36

Specialized in computer vision and deep learning, with research on diffusion models and efficient low-rank training of neural networks.

B.Tech — Electronics & Communication Engineering

Kalyani Government Engineering College

2018 – 2022 · CGPA 8.5

Certificate — Quantum Computing

Frontier Technology Institute

2020 – 2021

04

Research

2025

GILoRaT: Greedy Incremental Low Rank Training

Tamal Hansda, Samriddha Sanyal, Deborsi Basu

A greedy, incremental approach to low-rank training of neural networks — reducing parameter and compute cost while preserving model quality. Code available on GitHub.

Preprint · Vision & Intelligent Systems, E&ECE, IIT Kharagpur

05

Selected Projects

Open-source work — mostly things built from scratch to learn how they really work. See all on GitHub →

⚙️↗

SimpleGrad

An autograd engine built from scratch with a PyTorch-style API — layers, activations, Adam/SGD optimizers and loss functions, for educational clarity.

★ 6 · Python

👁️↗