Bhagyesh Kumar
@invi-bhagyesh

Bhagyesh Kumar(@invi-bhagyesh)

I am currently a sophomore in Department of Mathematics at MIT Manipal, majoring in Mathematics and Computing. I am working as a student researcher with Mars Rover Manipal(MRM) AI Research Team.


Previously, I worked on adversarial attacks for ocr models, and currently focusing on reasoning and alignment for small language models(SLMs). I'm also interested in AI Safety and hope to build a solid understanding of it in the future, since I'm still new to the area.

Selected Work

Computer Vision
TopoReformer: Mitigating Adversarial Attacks Using Topological Purification in OCR Models
Bhagyesh Kumar*, A S Aravinthakshan*, Akshat Satyanarayan*, Ishaan Gakhar, Ujjwal Verma
Accepted at AAAI Workshop, 2026
View Code → Paper →
LLMs
LM-quant

Benchmarking SLMs on quantization and RLHF effects to measure bias and fairness across standard evaluation datasets.

View Code →
LLMs
LM-collusion

Simulation of LLM agent collusion under Cournot competition strategies.

View Code →

How I fail

  • Nov 2025, AAAI Undergraduate Consortium Proposal not accepted
  • Oct 2025, Toporeformer(v1) rejected from NeurIPS Workshop 2025

Recent highlights

  • Nov 2025, Received $1000 scholarship to present my work at AAAI
  • Nov 2025, TopoReformer(v2) accepted at AAAI Workshop 2026