Headline

Associate Technical Architect at KPIT | Computer Vision, Multimodal Perception & ML Systems | Autonomous Driving and Agent Systems

About

I build AI systems that have to work under real constraints. At KPIT, I have spent the past seven years across autonomous driving, in-cabin intelligence and multimodal perception, progressing from hands-on development and inference optimization to technical leadership and architecture. My work spans computer vision, deep learning, multi-sensor pipelines and embedded deployment on NVIDIA and Qualcomm platforms, where model quality must be balanced against latency, memory, throughput and safety. My current role also includes advising teams across accounts on agent systems for engineering productivity, from use-case selection and evaluation to orchestration, tool integration, state and memory, and human approval. This is an extension of the same systems discipline: understand failure modes, measure behavior and turn research prototypes into dependable tools. I remain close to implementation through design reviews, prototyping, performance debugging and open-source projects. I am particularly interested in multimodal AI, efficient inference and dependable agent systems, and I write publicly about what I am building and learning. I am pursuing an M.S. in Computer Science at the University of Colorado Boulder.

Location

Pune District

AI stack

antigravity

claude code

codex

hermes agent

opencode

claudecode

Work experience

Associate Technical Architect at KPIT, Apr 2026 - Present · 5 mos, Pune District, Maharashtra, India, • Own architecture and technical direction for AI systems across autonomous driving, in-cabin intelligence, multimodal perception and internal agent initiatives. • Lead architecture reviews and mentor technical leads across 15–20-engineer programs, making system-level trade-offs across model performance, latency, compute, reliability and safety. • Guide research prototypes into deployable ML systems and define reusable patterns for integration, evaluation, deployment and observability. • Advise teams across accounts on agent use-case selection, feasibility and production design, including orchestration, state and memory, tool integration and human approval.

Sr. Technical Lead at KPIT, Oct 2022 - Mar 2026 · 3 yrs 6 mos, Pune District, Maharashtra, India, • Led technical delivery for two automotive R&D programs, guiding teams of 14–20 engineers across in-cabin intelligence and Level 3 autonomous-driving perception. • Designed multimodal perception pipelines spanning camera, LiDAR, radar, pose, gaze and driver-state models, balancing model quality with latency, memory, throughput and safety constraints on embedded compute. • Directed deployment and optimization of multi-model pipelines on Qualcomm Snapdragon Ride using quantization, knowledge distillation, graph optimization and performance profiling. • Stayed hands-on in design reviews, prototyping, performance debugging and system hardening while mentoring engineers and working directly with customer engineering teams.

Technical Lead at KPIT, Apr 2021 - Sep 2022 · 1 yr 6 mos, Tokyo, Tokyo, Japan, • Owned technical design and integration of a real-time lane and object-detection stack on NVIDIA Drive AGX. • Built high-performance middleware to schedule and execute proprietary perception models concurrently on embedded automotive compute. • Implemented a custom CenterNet pipeline in ONNX and optimized full-resolution inference to 33 FPS.

Freelance Software Engineer at Freelance Web Development, Feb 2017 - Dec 2018 · 1 yr 11 mos, Bangalore, • Neural Style Transfer: Developed a deep learning algorithm to generate artistic portraits, enhancing the visual appeal and user engagement of a mobile application. • Full-Stack Web Development: Designed and implemented a full-stack website for a medical billing company, integrating social features to improve user interaction and streamline operations.

Software Engineer at Accenture, May 2015 - Jan 2017 · 1 yr 9 mos, Bengaluru, Karnataka, India, •Developed scripts for automated usage processing for Data, SMS and MMS to facilitate testing. •Successfully migrated from batch billing(Amdocs Charging) to real time billing(SAP Convergent Charging). •Successfully upgraded Telus back-end Workflows from Mediation Zone 5 and Mediation Zone 6 to Mediation Zone 7 •Awarded ACE-Numero Uno Award for Q2 in FY2015

Associate Software Engineer at Accenture, Jan 2014 - Apr 2015 · 1 yr 4 mos, Bengaluru, Karnataka, India, •Service and development of Self Serve and Smart Desktop applications for Telus. •Service and development of SAP Convergent Charging to introduce real time rating for Telus. •Introduced 4G internet across Telus.

Education

Master of Science - MS, Computer Science, University of Colorado Boulder

Executive Post Graduate Programme, Machine Learning and Artificial Intelligence, International Institute of Information Technology Bangalore

Public links

github: https://github.com/himanshudongre