AI & Robotics Systems Engineer
I turn research-grade ideas into production systems that hold up under load. Papers on the theory, hands on the hardware.
01 / Profile
Columbia University MS · ROAR Lab 
Carnegie Mellon Research Scholar 
BITS Pilani Mechanical Engineering I take hard problems from proof to production — convex optimisation and learned policies on one side, lock-free C++ and real hardware on the other. Most engineers pick a side; the interesting work is the seam.
Mechanical engineering at BITS Pilani, research under Howie Choset at Carnegie Mellon, an MS at Columbia's ROAR Lab, and surgical robotics at Neocis in Miami. Four institutions, one thread: I want to know why a system behaves the way it does, and then I want to make it behave.
Which means the math is never decoration. A planner is only as good as the C++ underneath it, a model is only as good as the system ID behind it, and neither counts until it holds on hardware someone is paying for. I build all three layers.
At Armatrix, I lead a team of 5 building a hierarchical 3-layer control stack running at 20 Hz - cutting tracking error by 96% and MPC solve time by 82% for safe navigation at sub-5 cm clearance.
At Neocis, I developed dynamic models for the YOMI surgical arm that directly enabled advanced compliant motion and reduced system costs by 20%.
At CMU, I engineered a novel Heuristic-Search algorithm for pathfinding that outperformed baselines by 2-8x, resulting in IEEE RA-L publications.
From securing funding for bio-inspired underwater vehicles at BITS Pilani to designing haptic controls for rehabilitation at Columbia's ROAR Lab, I own the entire lifecycle, from first-principles math to hardware deployment.
Core competencies
MPC with recursive feasibility via convex decomposition. Deterministic 24-DOF SE(3) planning with Sobol-sampled D-PRM*. Solvers that hold their deadline.
Lock-free C++ at a hard 20 Hz - zero-copy POSIX shared memory, seqlock reads, and compile-time schema validation across a Python bridge.
Transformer-based MPC, RL policies, Gaussian-process system ID. Trained in MuJoCo and PyBullet, validated against instrumented hardware - not a leaderboard.
Team of 5, first-principles derivation through deployment. Funding secured, papers published, hardware in the field.
02 / Case study
Two builds carried end to end — from first-principles modelling to instrumented hardware.
01
Comprehensive, AI-powered web platform fundamentally accelerating chemical synthesis research. Integrating literature analysis, hybrid synthesis planning, reaction optimization, and logistical cost analysis into a cohesive workflow blending cheminformatics with deep learning.
02
Bio-mimetic underwater robot achieving locomotion through a stingray-inspired undulating fin mechanism. Engineered using Arduino/servos with a 3x3 factorial experiment in a water tunnel, quantifying thrust using NI DAQ and load cells for efficient underwater propulsion.
03 / Trajectory
Scrub the rail. Each stop is a system I owned, and what it cost to make it work.
June 2022 - July 2022
Pacify Medical Technologies
At Pacify Medical Technologies, I led the design of a compact syringe-gun injection mechanism. Using SolidWorks, I engineered a new holder and nozzle assembly that achieved a 25% reduction in size. To enhance device reliability, I integrated a force-sensitive resistor with an Arduino to provide active load measurement, effectively preventing breakage under operational stress. My role also involved collaborating with suppliers to source materials, which successfully reduced a potential budget overrun by 5%.
Mar 2023 - Dec 2023
Carnegie Mellon University
During my research at Carnegie Mellon University, I architected and implemented a novel Heuristic-Search algorithm in C++ to solve resource-constrained shortest path problems. By incorporating advanced techniques like dominance criteria and partial path pruning, my solution achieved a 2-8x performance increase over the existing baseline. Moreover, I also worked on testing algorithms developed by a colleague and to validate these findings in a real-world context, I developed a ROS2 framework to benchmark my algorithm against others, proving its superior efficiency in autonomous systems.
May 2025 - Aug 2025
Neocis Inc.
As a Robotics Intern at Neocis, I engineered an analytical dynamic model for the 7-DOF YOMI surgical robot arm, leveraging RNEA, CRBA, and System ID techniques. This high-fidelity model was instrumental in implementing Computed Torque control, which enabled advanced compliant behaviors. My contributions directly supported a strategic initiative to reduce the YOMI's production cost by 20%. I also designed critical CAD models for PCB fixtures to integrate new encoders into the arm's control system.
Bringing learnings from the USA back home
Following my time at Columbia and CMU, I've chosen to return to India to focus on deep-tech robotics. For me, this move is about prioritizing engineering substance over geography. My experiences in the US, from developing surgical robotics to engineering advanced pathfinding algorithms, honed my ability to bridge first-principles math with real-world hardware deployment. I'm excited to bring that technical rigor home to solve complex problems and contribute to the growth of India's robotics ecosystem.
Jan 2026 - Present
ARMATRIX · Full-time
At ARMATRIX, I lead a team of 5 engineers, architecting the end-to-end software stack for the robotic arm. I designed a hierarchical 3-layer control stack running at 20 Hz that cut tracking error by 96% and MPC solve time by 82%, with recursive feasibility guaranteed via convex decomposition and geometric refinement for safe navigation at sub-5 cm clearance. I developed deterministic 24-DOF SE(3) path planning with real-time collision checking using Sobol-sampled D-PRM*, and architected a zero-copy POSIX shared-memory transport layer, lock-free seqlock reads for the C++ real-time loops with a semaphore-blocked Python bridge, hardened by compile-time schema validation to eliminate silent memory corruption. I also engineered a high-fidelity MuJoCo simulation environment, bridging the Sim2Real gap through physics modeling and system identification.
04 / Index
Control stacks, learned policies, multi-agent systems, and a few weekends that got out of hand.
06 / Contact
If you're building something that has to work outside a simulator - or you just want to argue about controllers - I read every email.