I build systems that are measurable rather than merely impressive — equation discovery for physical transport, explanations you can falsify, and research infrastructure that keeps its own record.
The through-line across everything below is one question: can this system show its work, and does the account it gives actually hold?
Research & publications — text version
| Since | Role | Where |
|---|---|---|
| Aug 2026 | Research Assistant — Autonomous Research Supervision Agent | Anna University · supervised by an IBM India Enterprise Architect |
| Feb 2026 | Undergraduate Researcher — Symbolic Regression for Packed-Bed Transport | IIT Madras · Advisor: Dr. Himanshu Goyal |
| — | Research Engineer — AI Research Infrastructure | University of Toronto (remote) · Collaborator: Tong Li |
| Apr 2025 | Undergraduate Researcher — Sign-Language Translation & Explainable Medical Imaging | Anna University, MIT Campus |
Neural Translation of Tamil to Indian Sign Language via Pivot-Transformer with 2D and 3D Avatar Motion Through English as a Medium Language
Radha Senthilkumar (Supervisor) · Ruparagunath G · Jayanathi P (Mentor) — ICIETSD 2026 · IEEE Xplore
From ASL to ISL: Translating Gestures across Sign Languages
Co-author — ICCCSP 2026 · Springer (IFIP-endorsed) · Accepted, to appear
Synapse — Does a retrieval system's explanation actually cause its ranking?
A counterfactual edge-ablation protocol: delete exactly the edges an explanation cited, re-rank, and test the displacement against a count-matched random-edge control. Personalized PageRank written from scratch so per-arc flow is recorded, not reconstructed. 215 tests · Python
NeuroLens — Faithful XAI for multiple sclerosis screening
Explanations gated on whether the model actually decided inside the interpretable basis, rather than assumed to be faithful.
Universal Pressure-Drop Correlation — Scientific ML · equation discovery
Symbolic regression recovering closed-form transport correlations for packed beds.
MS vs. Cerebral Small Vessel Disease — Bayesian ML under scarce clinical data
Distinguishing MS from CSVD on MRI with severely limited data — 20 of 21 patient cases correctly classified.
Computer Vision Lab — UG semester-4 coursework, kept public as a working record.
Several projects above are research in progress or live in private repositories; the portfolio carries the full write-ups.
B.Tech, Artificial Intelligence & Data Science — Anna University, MIT Campus · Chennai
Diploma — Central Polytechnic College · Chennai