My projects

Personal & academic projects I worked on.


AI Thermometer

Contactless Temperature Measurement with IR & RGB Cameras

This project uses Facial Detection and radiometric IR measurement in order to determine person’s facial temperature. Hardware and software are freely available on github. Highlights:

  • DNN-based Facial Detection with RGB camera.
  • Contactless temperature measurement with IR camera.
  • Option for auto-calibration with external reference (blackbody).
  • The project was awarded as Project of the Month by Nvidia.
  • I presented AI Thermometer during a GTC’21 session
Star Fork Watch

Wildfire Detecting Drone


In this project I implemented and deployed DNN-based wildfire detection with RGB camera.

  • Trained DNN for fire detection
  • Scrapped and hand annotated dataset of 10k fire scenes
  • Trained model to achieve 94.7% accuracy and 98.7% sensitivity
  • Deployed DNN on Nvidia Jetson Drone
  • Tested over an 80-acre wildfire with Sacramento Metro Fire Dept
Star Fork Watch

Wildfire Simulation


A project simulating and visualizing the dynamic behavior of a wildfire. Built as an evironment to train RL agents.

  • Simulates the effect of wind direction on the spread
  • Can run with custom fuel distribution
  • Adapted the approach from this paper
  • Optimized the approach from \( O(n^4) \rightarrow O(n \cdot log(n) \)
  • Runs on 10M cell lattice at \( 20FPS \)
  • I’m working on scaling this approach further with CUDA.
Star Fork Watch

Autonomous Vehicle Camera + Radar Perception


In this project I retrofitted a 1999 Mercedes CLK with:

  • RTK GNSS,
  • stereo cameras,
  • automotive radar
  • Nvidia Jetson Nano.

Then, I trained & implemented 2D perception (mRCNN-based), radar processing pipeline & image+radar fusion. As a last step I tested the system on the road. For more information go to related paper. On the right you can watch a short video demonstrating the mask-RCNN 2D perception on SJSU campus.

Star Fork Watch

Paper Search Engine

Search all AI papers on arxiv.org in \(\lt 1ms\)

The app rapidly compares user query against abstracts of all CV/AI/ML papers available on arxiv.org. I wrote it to ensure that someone didn’t come up with idea before. The app also allows you to find papers similar to your research interests.

  • A typical query takes 1ms to process.
  • Paste full abstract of an interesting paper to find similar papers.
  • You can search 30 years of AI research in a fraction of a second!
Star Fork Watch

PCB Business Card with Proto Board


A PCB businesss card with matt black soldermask looks and feels much better than a boring paper business card. It has a small prototyping area on the side, and makes a great converstaion starter! The schematic and board files are available on my github.

Star Fork Watch