Foad Namjoo

I’m a Computer Science Ph.D. student at the Kahlert School of Computing (University of Utah), advised by Jeff Phillips in the UtahDB Lab. I work on interpretable/controllable LLMs, ML, high-dimensional data geometry, and reliable wearable data systems.

What I build

• Activation steering for LLMs — residual-stream injections (layer/α sweeps) to control verbosity & tone in models like Llama 3.1 and Qwen-2.5, plus guardrailed evaluation (LLM judges + JSON schema) to reduce hallucinations.
  

• dKS test (efficient & stable) — near-linear (O(n \log n)) 2D two-sample test with ε-accurate, unit-invariant IPM; improved stability vs. mdKS. arXiv:2504.11299

• MotionPI — a privacy-first wearable sensing platform (Flutter + BLE wristbands → Node.js/Express + MongoDB).
  Offline-first sync, schema-validated writes, and reliability guards (auto-reconnect, buffering, schedules). Handled ~7.7M records/day with 0 malformed writes. Worked with a multidisciplinary group in Health & Kinesiology Sciences and the Huntsman Cancer Institute, and with hardware collaborators at The Ohio State University; led cross-functional teamwork (Utah ↔︎ Ohio State) to deliver reliable field data collection.
  SmartSP 2025 paper (accepted)

Recent

• Poster: Presented dKS research at the Utah AI Summit 2025 (University of Utah) — June 18, 2025.
• Designing a Secure and Resilient Distributed Smartphone Participant Data Collection System. EAI SmartSP 2025 (accepted).
  Publications page · PDF
• Efficient and Stable Multi-Dimensional KS Distance. arXiv 2025.
  arXiv · PDF
• HCD-Net for Hyperspectral Change Detection. Remote Sensing 2024.
  See all