yash khurana

i'm yash. i'm an ai research engineer interested in edge inference, hyperdimensional computing, and perception that survives contact with the real world. i'm currently at arch systems, building low-power multi-sensor detection.

i hold a masters in artificial intelligence from northwestern and a bachelors in computer science from vit.

off the clock, i'm building overtchat — a self-hosted chatgpt: same experience, none of the surveillance. local models, no subscription, nothing you type leaves the box. coding agents too.

Portrait of Yash Khurana at dusk, city skyline behind
fig. 00 — subject // rgb + ir, misregistered

experience

AI Research Engineer, Arch Systems LLC · SDCP (AFRL)Nov 2025 — present

  • Detects objects from fused visible + thermal images (single FPGA, no CPU).
  • Beats the GPU on energy by 3.4× — 21.8 W inside a hard 25 W budget.
  • Runs on hyperdimensional computing — AI that bend under noise and attack instead of breaking.
  • The fusion beats published sota — +4.5 mAP average, +12.7 after dark.

AI Systems Engineer, Intel Labs · AFRL HDC InitiativeJan 2025 — Oct 2025

  • Lead engineer on an Air Force program to make chips learn on-device.
  • Taught a detector to adapt on the fly — VSA encoders that fit the learning path into 0.5 MB of silicon.
  • Built the scoreboard: automated accuracy/latency/power benchmarks from server GPUs to microcontroller FPGAs, on MLPerf methodology.

Graduate AI Research Intern, Intel LabsJune 2024 — Aug 2024

  • Shrank an object detector to a fifth of its size — 90% faster on CPU, under 5% mAP loss.
  • Swapped the classifier for a hyperdimensional one — 20% smaller again, and tougher against noise and adversarial attacks.

education

MS, Artificial Intelligence, Northwestern UniversitySept 2023 — Dec 2024

  • GPA 3.85 / 4.0

B.Tech, Computer Science, Vellore Institute of TechnologyJuly 2019 — June 2023

  • GPA 8.79 / 10

research

Radiation Oncology, Feinberg School of Medicine2024

  • RAG Q&A for patients with citations that check out.
  • Fine-tuned Llama-3 into a rad-onc tutor.
  • Eval metrics that actually track human judgment.

IVP Lab, Northwestern University2023 — 2024

  • Self-supervised XRF denoising of old-master paintings.
  • Taught an RL agent to route the XRF scanner.