Headline

Computer Vision & Applied ML

About

Over 8 years I've worked across automotive safety (Mercedes-Benz R&D), XR perception (Qualcomm), and now insurance-scale damage assessment (Tractable) — always at the point where computer vision meets a real decision: is this calibration drifting, is this damage segmentation good enough to ship, is this experiment worth the compute. Recent work: kill-gated experiment design and bootstrap-based significance testing for a production RF-DETR segmentation model across 650+ scans, fine-tuning ViT-based damage detectors (+8% recall on high-severity collisions), and calibration pipelines that cut XR perception validation time by 40%. I'm currently exploring Applied and Data Scientist and Computer Vision roles where rigorous experimentation and production ML meet — open to a conversation if that's you too.

Location

Greater Bengaluru Area

AI stack

claude code

claude

notebooklm

huggingface

ChatGPT

Microsoft Copilot

Work experience

Data Scientist at Tractable, Apr 2026 - Jul 2026 · 4 mos, Bengaluru, • Fine-tuned a ViT-based damage-detection model on multi-camera vehicle imagery, improving recall by 8% on large-collision cases through targeted data ingestion • Drove a staged, kill-gated experimental pipeline investigating damage-segmentation scoring for a production RF-DETR model, testing hypotheses cheaply before committing to expensive end-to-end runs • Built a panel-level evaluation harness and ran bootstrap-based significance testing across 650+ scans, delivering a go/no-go verdict and surfacing an actionable lever (per-class thresholding,∼15% relative F1 headroom) • Migrated training-data pipelines to a centralised, quality-vetted dataset store, documenting legacy data-farming methods and their warehouse equivalents

Senior Engineer at Qualcomm, Mar 2023 - Apr 2026 · 3 yrs 2 mos, Bengaluru, • Designed and deployed camera–IMU calibration analysis pipelines to evaluate intrinsic and extrinsic stability across large-scale XR sensor deployments, improving calibration anomaly detection by 30% • Built automated calibration and point cloud processing pipelines with CUDA-accelerated GPU execution and CI/CD integration, enabling repeatable geometric validation and reducing end-to-end processing time by 40%. • Owned calibration validation, 3D reconstruction accuracy metrics, and performance KPIs for production XR perception systems using UR10e and EPSON robotic arms. • Built telemetry dashboards in Power BI enabling faster regression analysis and cutting decision time by 20% for engineering leadership • Proposed and built an AR Navigation app for visually impaired users using real-time object detection and depth estimation

Computer Vision Engineer at Mercedes-Benz Research and Development India, Aug 2021 - Mar 2023 · 1 y 7 mo, Bengaluru, Karnataka, India, • Built multi-camera calibration solution achieving sub-mm accuracy for Driver Monitoring Systems. • Designed depth generation pipeline using 3D transforms, improving spatial accuracy by 15%. • Established dataset QA protocols (99%+ precision) for safety-critical recognition systems. • Delivered eye-gaze analytics pipeline using MTCNN + embeddings; presented to Mercedes-Benz and Nike CXOs at ADAS workshops. • Collaborated onsite with teams in Germany and Spain to drive data-collection campaigns and align KPIs across engineering groups.

Associate at PwC India, Jul 2016 - Jul 2019 · 3 y, Gurgaon

Education

Master of Management, Business Analytics, Indian Institute of Science (IISc)

Bachelor of Technology (B.Tech.), Computer Science, amity University

Public links

github: https://github.com/abhi5454