I build AI systems that perceive, understand, reason, plan, and act over long horizons in the physical world.
My work spans LLMs and agentic systems, computer vision, and reinforcement learning, from product development and production ML & data platforms at Uber to peer-reviewed conference papers at the IEEE Int. Conf. on Robotic Computing and Springer CCIS, with workshop papers at CVPR and AAAI. With that foundation, I'm now moving toward robot learning and embodied AI, where reinforcement learning, world models, and multimodal foundation models meet perception, planning, and control on real robots.
I develop AI systems that perceive, reason, plan, and act over extended horizons in the physical world.
Agents that perceive, learn, and act in the real world, bridging sim-to-real RL and language-conditioned policies.
Hierarchical, adaptive planning for tasks that unfold over time.
Learning predictive models of dynamics to simulate and imagine.
Symbolic and neural reasoning for reliable decision-making.
Grounding language, vision, and action, toward vision-language-action policies for control.
AI that generates hypotheses, designs experiments, and verifies.
Research spanning reinforcement learning, computer vision, and AI for scientific discovery.
Led the backend re-architecture of the promotions/discount platform for stackable cart offers across 4+ partner teams, and shipped a unified diagnostic API at 99.99% uptime and P99 < 30 ms.
Built idempotent, fault-tolerant reporting APIs that scaled monthly statements to ~15M organizations at ~1400 RPS with 99.99% availability, plus a large-scale validation framework and workflow-orchestrated data pipelines.
Built Redis-backed RPC APIs that cut query time from 250 ms to 10 ms, a tree-based dynamic form-rendering system, and a Graphviz form-flow visualizer; cut storage 92.6% by shrinking a join from 175M to 13M rows.
Built a LightFM-based hybrid video recommender on 1M+ records, plus Node.js/MongoDB APIs and SQL/BigQuery pipelines driving 28%+ engagement and 12%+ retention gains.
The languages, frameworks, and platforms I reach for, from research prototypes to production systems at scale.
19 honours across international and national competitions, spanning Inter-IIT Tech Meet golds, the World Programming Championship, and national olympiad rankings.
A five-year dual degree in Computer Science and Engineering at IIT Kharagpur, on a foundation built at Apeejay School, Nerul.
A path shaped by curiosity, learning, and a drive to build intelligent systems that matter.
Identity-preserving image-to-video generation via vision-grounded prompting.
Fault-tolerant distributed consensus built on the Raft algorithm.
A lightweight Qwen-style transformer, implemented from scratch.
Probabilistic reliability scoring for honest evaluation of AI agents.
A minimalistic file system with modern read and write capabilities, in C.
An Electron desktop app for smooth live screen recording.
I'm always excited to connect with researchers, collaborators, and curious minds shaping the future.
Get In Touch →