I benchmark AI on real hardware.
I'm an AI/ML researcher and software engineer making AI systems faster, more reliable, and ready for the real world. Right now I'm measuring how vision-language models hold up on the NVIDIA DGX Spark under sustained load.
// about
I'm a Computer Science senior at the University of North Texas with a minor in Physics. My path started with my hands inside computers and robots in Nepal, and it's led me to research how large AI models actually perform on the machines they run on.
I care about AI that holds up outside the demo: fast under real workloads, reliable in what it says, and useful in the physical world. That's the thread connecting my research, the LLM apps I build, and where I'm heading next with quantum machine learning and robotics.
The areas I'm building expertise in, now and next.
// research
Most benchmarks test a cool machine for a few minutes. Real deployments run for hours. My research measures what happens in between.
NVIDIA-sponsored, Responsible AI Lab, Sept. 2026 to present
How does sustained thermal load affect vision-language model inference on the NVIDIA DGX Spark?
I'm designing the study and building a Python harness that lines up every request's latency with what the GPU was doing at that exact second. The harness will be released as open source, so anyone with the hardware can reproduce the results.
// projects
From LLM applications to databases to robots, each one solving a real problem end to end.
LLM application, team project
Practice interviews that feel real. HireMate asks role-specific questions, evaluates every answer with Claude and OpenAI models, and gives feedback specific enough to act on, then tracks your progress over time.
My role: Project lead. I led the team, designed the prompt workflows, and refined model outputs to improve feedback quality.
Robotics, embedded systems
A prototype that seeds, waters, and monitors soil moisture on its own, with a C++ app for remote control. Made reliable through repeated testing and troubleshooting.
Database systems
A relational database that turns raw federal court records into query-ready data, designed from ER model through stored procedures, triggers, and complex SQL.
Data structures, performance
A driver records system built on custom hash tables and linked lists for constant-time average lookups, with categorization, license tracking, and record migration.
Senior capstone, in progress
The year-long capstone my team, Visionary Creators, is building at UNT. More details coming soon.
// journey
Diagnosed and repaired computers. My first real look inside the machine.
Assistant project designer on robotics and automation projects.
Physics minor, graduating May 2027. Led full-stack and database team projects.
Undergraduate researcher on an NVIDIA-sponsored inference study.
Specializing in AI/ML, with quantum machine learning and robotics ahead.
// skills
// contact
I'm open to AI/ML research collaborations and internships. The fastest way to reach me is email.