Experience

A timeline of my work, internships, and professional journey.

Research Intern
Dec 2025 – Present
UTRGV — AIM Lab × VIT Vellore · Remote / Vellore
Physics-informed & uncertainty-aware ML for structural materials — bendable concrete and crack assessment.
ECC (preprint)ECC-pipeline (repo)PINN (preprint)Crack CV + PINN (repo)
the full story, bullet by bulletExpandCollapse
Built an end-to-end pipeline predicting the tensile properties of Engineered Cementitious Composites (“bendable concrete”) — strength and ductility — from mix design, with calibrated 80% prediction intervals via CatBoost + Mondrian Conformal Quantile Regression.

Developed a visual crack-inspection pipeline: YOLO-seg extracts per-crack geometry, then a physics-informed neural network solves for the Mode-I stress-intensity factor (K₁) and returns a severity verdict against fracture toughness.

Treated the PINN as an offline solver (trained once per a/W ratio via LEFM similarity), so every detected crack is evaluated in microseconds at inference time.
Robotics Intern
Aug 2025 – Jan 2026
Kyungpook National University (KNU) — PIAI Lab × VIT Vellore · Remote / Vellore
Coverage-planning & gas-source localization for autonomous robots, under Dr. Yuvaraj.
HazMap (preprint)HazPatrol (repo)Gas-source localization (repo)
the full story, bullet by bulletExpandCollapse
Worked under Dr. Yuvaraj on autonomous-robotics research (KNU PIAI Lab × VIT Vellore collaboration).

Built HazPatrol — autonomous coverage of unknown areas using a modified C* algorithm that incrementally builds a Reachability Connectivity Graph (RCG) and drives boustrophedon laps until the full navigable area is covered.

Implemented a ROS 2 gas-source localization stack (tested in GADEN) with CUSUM-based hotspot detection, real-time concentration mapping, and Bayesian source estimation — used to produce the results for the paper.
Research Intern
Oct 2024 – Jun 2025
Samsung Research, Bangalore · Remote
BQ2Viz — a multi-agent GenAI framework turning natural language into data visualizations.
NL → Data Viz (repo)
the full story, bullet by bulletExpandCollapse
Engineered BQ2Viz, a multi-agent GenAI framework (LangGraph, LangChain, ChromaDB) that converts natural language into data visualizations, using a Profiling Agent for automated schema indexing and a Self-Correcting Code Generator.

Designed a Reasoning-Based Planner for multi-file retrieval, inferring multiple distinct visualizations from a single abstract query (e.g., “Analyze Sales” → trend line + category bar + regional pie).

Benchmarked against NVAGENT on the VisEval dataset: 84.57% accuracy on complex multi-table queries, cutting the error rate to 11.5% via a feedback-driven code-repair loop.
Software Engineer — Autonomous Domain
Apr 2024 – Present
Team Vyadh (Mars Rover Team), VIT · Vellore
Autonomous navigation, perception & state estimation for a Mars rover — 13th at IRC, 5th at IRDC.
Obstacle & pit avoidance (repo)Rover stack (repo)
the full story, bullet by bulletExpandCollapse
Apr 2024Joined the autonomous domain teamJan 2025Promoted to Senior Autonomous Member ↑
Engineered the autonomous navigation stack with ROS 2 (Nav2) and SLAM, implementing a Frontier Exploration algorithm to map unknown environments within dynamic GPS geofences.

Built a real-time obstacle & pit avoidance perception pipeline using Intel RealSense D455 + PCL, with odometry-based path recovery and a custom PID alignment to compensate for wheel drift on uneven terrain.

Fused IMU + odometry with an Extended Kalman Filter to cut state-estimation noise, and built a telemetry GUI for Astro-Bio sensor streams (CO₂, IR, moisture).

Part of the team that placed 13th at the International Rover Challenge (IRC) and 5th at the IRDC.

Research Intern

Robotics Intern

Research Intern

Software Engineer — Autonomous Domain