Headshot of Ankur Jai Sood

Contact Information

San Francisco, CA
ankur.jai.sood [at] gmail.com
LinkedIn GitHub Google Scholar ORCID

Ankur Jai Preferred name Sood

Senior software engineer building L3+ autonomy platforms, shipping reliable data pipelines and safety critical services that keep vehicles compliant and support perception, planning, and recording at scale.

Passionate About:

  • Public transit and walkable cities
  • Cars and driving on a great backroad
  • Cooking experiments
  • Games with strategy and storytelling (Catan, RimWorld, AoE II)
  • Long hikes with food-related rewards at the end

Exploring and Building:

  • AI safety and AI generated content detection
  • AI/ML systems for robotics & autonomy
  • Agentic AI, tool-using agents, and self-evolving agents

Present Positions

  • Senior Software Engineer Nvidia · Jul 2022 - Present
  • Advisor Wosler Corp · Jan 2022 - Present

Prior Experience

  • Senior Software Engineer Lucid Motors (Formerly: Atieva) · May 2022 - Jun 2022
  • Software Engineer Lucid Motors (Formerly: Atieva) · Jun 2020 - May 2022
  • Software Integration Engineer Ford Motor Company · Apr 2019 - Jun 2020
  • Application Engineer Bluewrist · May 2018 - Apr 2019
  • Development Engineering Associate (Intern) Celestica · May 2016 - Aug 2017
  • Research Student Hoffman Lab, University of Toronto · May 2015 - Jun 2017
  • Research Student STTARR, University Health Network · May 2014 - Aug 2014

Publications

DNAmod: the DNA modification database

Sood, A.J.*; Viner, C.*; Hoffman, M.M. · Journal of Cheminformatics, 2019

Patents & Patent Applications

Data Flow Management for Computational Loads

US12430067B2 · Atieva Inc

Status: Granted · Date: 2025-09-30

Lidar Data Compression and Processing

US20250044413A1 · Atieva Inc

Status: Pending · Date: 2022-10-27

System, Method and Device for Remote Ultrasonography

US20250268582A1 · Wosler Corp

Status: Pending · Date: 2023-04-18

Academic Projects

Eye in the Sky: Live Blackjack Card Counting via Real-Time Video Analysis

CS231N · Stanford · Jun 2025

  • Real-time pipelines combining YOLOv5 for detection/classification and SegFormer for segmentation under occlusion.
  • DeepSORT based tracking with deduplication heuristics to maintain persistent IDs and prevent double counting.

Settling the Odds: AI Agents in Catan

AA228/CS238 · Stanford · Dec 2024

  • Framed 4 player Catan as an offline RL problem with reproducible data, training, and evaluation pipeline.
  • Implemented DQN with soft target updates, Huber loss, and invalid action masking plus heuristic reward shaping.

Strategies for Optimization of minBERT for Multi-Task Learning

CS224N · Stanford · Jun 2024

  • Implemented multi-task minBERT with sentiment, paraphrase, and similarity heads, outperforming single task baselines.
  • Explored task schedulers, LoRA, and Projected Attention Layers to cut trainable parameters while improving adaptation.

Variable Compliance Instrument for Laparoscopy

Mechatronic Systems Engineering Capstone · University of Western Ontario · Jun 2018

  • Designed and developed a variably compliant laparoscopic instrument that warns operators near obstacles and softens to reduce patient risk.
  • Awarded 3rd place at the 2018 Mechatronic Systems Engineering Capstone Design Showcase.

Personal Projects

Lookout

AI marketplace research agent to help organize listings and make purchasing decisions. It scores listings, explains reasoning, asks smart follow up questions, and remembers your preferences across sessions.

Beeline

AI powered multimodal route planning for the Bay Area. Combining driving, parking, and transit in one optimized trip.

HALpdesk

Terminal copilot that turns natural language into bash commands. Provides a chat feature for guidance, juggles multiple sessions, and warns users before running dangerous commands.