Ankit Billa
Senior Perception Engineer · San Francisco
I build autonomous perception systems, turning raw sensor data into real-time scene understanding. At Sauron, I'm building an autonomous home-security system powered by the perception stack of an autonomous vehicle, to identify threats and send alerts in real-time. I primarily work at the intersection of deep learning, computer vision and machine learning.
Along the way I've been lucky to learn from brilliant mentors:
- Prof. Jianbo Shi (GRASP Lab, University of Pennsylvania)
- Prof. Jürgen Beyerer & Dr. Tim Zander (KIT & Fraunhofer IOSB)
M.S.E, Computer & Information Science
University of Pennsylvania
B.Tech, Computer Science & Engineering
Punjab Engineering College, Chandigarh
Senior Perception Engineer
San Francisco · May 2025 – PresentWork
Multi-camera tracking and re-identification built for confined, high-traffic spaces.
Research
Academic and applied research across depth, motion, material and language.
Off-Road Terrain Classification. Depth-assisted semantic segmentation on a novel RGB-D wild-terrain dataset.
Egocentric Interaction Tracking. Hand–object interaction detection grounded in 3D from egocentric video.
Paper Fingerprinting. β-Variational-Autoencoder embeddings that fingerprint paper textures to fight document fraud.
3D Scene-Flow Mesh Tracking. Projecting a rigid 3D mesh onto a moving subject via scene flow and pose.
Projects
Engineering and research across systems, ML, graphics and bio-engineering.
Mini Minecraft. A from-scratch C++ voxel open world with procedural terrain, biomes and NPC AI.
PennCloud. A fault-tolerant distributed clone of Gmail + Drive with strong consistency.
StreamWorks Search Engine. Crawler, indexer and ranker over a 500k-URL corpus.
Finger Flexions from ECoG. Decoding intracranial brain signals into continuous finger movement.
Counting Occluded Parts. Density-map FamNet hitting 1.96 MAE on highly-occluded washer parts.
Video Virality Prediction. A multimodal Video-Vision Transformer reaching 82% accuracy.
Autonomous Trajectory Planning. A Deep Q-Network driving agent in the CARLA simulator.
Text-to-Face Generation. StackGAN + ProgressiveGAN portraits synthesised from text descriptions.
NarcoSoft. An Android platform built for the Government of Punjab's drug-abuse-prevention program.
You can reach out to me via the email below.