Researcher first. Engineer next. Always a student.
Third-year CS + AI. I read the papers, then I ship the system.
I go deep before I build. When I pick up a problem I read the papers, trace the baselines, and find the gap. My PM2.5 forecasting work started with understanding why standard temporal models lose recall on extreme pollution events — the architecture came from that analysis, not the other way around. That research-first habit runs through everything I work on.
On the engineering side I close the gap between prototype and production. The SRE Incident Responder ran against real Redis and PostgreSQL with live state graders, not mocks. ARM-Gym put every generated assembly through three-stage hardware verification. I care about the distance between a working demo and a system that holds up under real conditions, and I close it deliberately.
Hover to see relationships
Selected from online qualifiers. 36-hour on-campus hackathon hosted by IIIT Hyderabad, IIT Delhi and IBM. Pitched a working ResGRU-UNet prototype to industry judges. Top 4 from 10 finalist teams.
IIIT-H × IIT Delhi × IBM pollution forecasting competition. Ranked #5 out of 77 teams to advance to the offline round.
Built an OpenEnv-compatible SRE incident responder simulating a production microservice cluster with real Redis and PostgreSQL. Selected as finalist.
NVIDIA Nemotron Challenge (#1427 / 3,252), Orbit Wars — ongoing.
Contributing speech-recognition and intent-classification features for a voice-controlled home automation platform. Designing the NLP pipeline to map spoken commands to structured device actions.
Merged 6+ pull requests across open-source repositories. Contributions included bug fixes, code quality improvements, and documentation updates across multiple codebases.
Ranked #1427 / 3,252 globally in the NVIDIA Nemotron large language model evaluation challenge.
Currently open to internships and new-grad roles for 2026.