uni of nottingham · 2026 graduate in msc ai (2yr)

Veerendhra Kumar

MSc Computer Science graduate from the University of Nottingham with a strong interest in backend development, machine learning, and cloud computing. My academic work focused on medical AI and I co-authored a research paper on ECG classification using deep learning and explainability techniques. Alongside research, I enjoy building practical systems — I'm currently working on the backend of an Allergens Manager, designing the data models, core logic, and APIs behind it. I like solving real problems with clean, reliable code, and I'm always keen to keep learning; right now I'm building my skills in cloud (AWS) and ML engineering. Open to opportunities where I can grow as a developer and contribute to meaningful projects.

say Hi to TARS!!!

01 · about

About

click HERE :]

I love building things that cannot be built. Love reading, researching, and exploring different topics in different ways :]

Right now I'm working on the allergens ops backend for Nottingham Venues, and spending my free time on LLMs, automated AI bots and programming. Looking for chances to build things people actually use.

building with

python AI frameworks LLMs RAG robotics Git

currently learning

Transformers tokenizers Obsidian
02 · projects

Projects

python · grad-cam · ai

Explainability as a Training-Time Reliability Signal for Efficient ECG Classification

Training deep models on 12-lead ECGs is expensive, and a lot of that cost is wasted on samples the model has already mastered. This repo asks a simple question: instead of throwing data away based only on loss or confidence, what if we also ask where the model is looking? We use Grad-CAM as a cheap, per-sample reliability signal during training and drop the samples that are either already easy or that the model is “getting right for the wrong reasons.” The result is a family of progressive data-dropout schedules — with and without the explainability signal — that cut the number of samples actually backpropagated each epoch while trying to hold onto macro-F1. It runs across three public ECG corpora (PTB-XL, CPSC 2018, Georgia 2020) and three backbones (EfficientNetV2-S, ResNet-18, MobileNetV2), and logs everything you need to compare runs: accuracy, macro-F1, effective epochs, kept-ratio curves, and Grad-CAM figures.

python · automation

career oops

Searches for jobs against your requirements and works efficiently by tailoring a fresh CV for every role — 100% ATS-friendly. Still under process…

upcoming ideas

house finder

Not yet started.

03 · experience

Experience & Education

  • 2020 — 2024

    B.E in Computer Science — Sathyabama University

    Graduated with 8.5 CGPA (First Class). Worked on NLP and automated CV analysis for the final year project.

  • 2024 — 2026

    MSc in Computer Science with AI (2 yr) — University of Nottingham

    Secured a distinction in my dissertation. Worked on efficient ML and DL models and researched XAI for efficient AI-enabled models on medical time-series data. Journal under review:

    arxiv.org/abs/2606.12252 →
  • 2026

    Conference Assistant & Backend Developer — Nottingham Venues

    Working on the allergens ops backend :)

04 · music

I love music

now spinning — tap the record to listen with me

I love music — it's always on while I build, and it's half the reason anything on this page moves. The record links straight to what I've had on repeat lately.

suggest me a song

no account needed — it opens your mail app and the song lands straight in my inbox

05 · scribble

Leave a scribble

Suggest a song, doodle something, write me a note — anything goes. Draw it below and hit send; it reaches me as a picture.

sends as an image — on phones it opens your share sheet, on desktop it downloads and opens your mail

06 · contact

Get in touch

I actually reply. Whether it's an internship, a project, or just feedback on this page:

veerendhrakumar0@gmail.com
Full graduation photo — Veerendhra Kumar in front of the university clock tower