Eesh Saxena
Engineer + AI Researcher, CP Programmer, ML Engineer, NLP Engineer, and Full-Stack Dev
Hi there
I'm Eesh, a CS Engineering student at IIIT Senapati, Manipur. I'm a Research Intern at IIIT Vadodara working on Reversible Data Hiding in Encrypted Images, and previously interned at IIT Tirupati's SEVA Lab on multi-object tracking using transformer-based approaches (MOTIP, CVPR 2025).
Outside research I spend a lot of time on competitive programming: Specialist on Codeforces (1582), Guardian on LeetCode (1873), and 4★ on CodeChef (1866). I've solved 1500+ problems and placed well in global contests. Want to talk research, internships, or AI? Get in touch.
Send me a messageSelected
Projects
Work across graph-based reasoning, AI agent security, and open-source tooling.
View all projects →Conflict-Aware Graph RAG
Graph-based RAG pipeline converting text to knowledge triples via LLMs. Entropy-based conflict module reduces hallucination on multi-hop QA.
LLM Agent Security
Attack-and-defense work on tool-using LLM agents: a clean-room benchmark for guardrail bypasses and confused-deputy data exfiltration, plus a published Kaggle working note on hardening agents against prompt injection.
Outreach Emails Toolkit
Open community database and a stdlib-only Python mail-merge toolkit for internship cold outreach: batched sends with dry-run-first safety, plus a career-page watcher that opens an alert when new roles are posted.
Research & Work
Research Intern
IIIT Vadodara · 2026 (6 months)
Under Dr. Abhisek Paul
Analysed Zhang (IEEE SPL 2011) on Reversible Data Hiding in Encrypted Images. Studied 30+ related works, implemented the encryption, embedding, and extraction pipeline in Python, and achieved 100% reconstruction fidelity with a 0.5 bpp embedding rate.
Winter Research Intern
IIT Tirupati, SEVA Lab · 2025 (Winter)
Under Dr. Chalavadi Vishnu
Reproduced MOTIP (CVPR 2025) for Multiple Object Tracking. Evaluated transformer-based identity prediction, achieving 68.2 MOTA and 64.5 HOTA on MOT17 datasets by optimizing query initialization and spatio-temporal embeddings for dense scenes.