Looks like you have JavaScript disabled. For the full Twine experience, you will need to re-enable it.

I’m Aman Shekh, an ML Engineer and technical project lead who builds production-grade computer vision systems across edge and cloud. I’ve spent 6+ years taking models from end-to-end development through embedded porting, backend services, and QA—often optimizing for real-time performance and robustness across multiple hardware tiers. At Motive, I lead end-to-end delivery for an in-cabin passenger detection system, including architectural decisions (like moving to full-resolution inference), DRA/discrepancy analysis to close gaps between on-device and offline results, and orchestration of ML/embedded/EFS pipelines. I also enjoy engineering practical tooling that reduces annotation cost and improves quality, such as using VLM-based event gating for human-in-the-loop workflows.…I’m Aman Shekh, an ML Engineer and technical project lead who builds production-grade computer vision systems across edge and cloud. I’ve spent 6+ years taking models from end-to-end development through embedded porting, backend services, and QA—often optimizing for real-time performance and robustness across multiple hardware tiers. At Motive, I lead end-to-end delivery for an in-cabin passenger detection system, including architectural decisions (like moving to full-resolution inference), DRA/discrepancy analysis to close gaps between on-device and offline results, and orchestration of ML/embedded/EFS pipelines. I also enjoy engineering practical tooling that reduces annotation cost and improves quality, such as using VLM-based event gating for human-in-the-loop workflows.

Aman Shekh

AI Engineer, Data Scientist, QA Engineer, +4





I’m Aman Shekh, an ML Engineer and technical project lead who builds production-grade computer vision systems across edge and cloud. I’ve spent 6+ years taking models from end-to-end development through embedded porting, backend services, and QA—often optimizing for real-time performance and robustness across multiple hardware tiers. At Motive, I lead end-to-end delivery for an in-cabin passenger detection system, including architectural decisions (like moving to full-resolution inference), DRA/discrepancy analysis to close gaps between on-device and offline results, and orchestration of ML/embedded/EFS pipelines. I also enjoy engineering practical tooling that reduces annotation cost and improves quality, such as using VLM-based event gating for human-in-the-loop workflows.…I’m Aman Shekh, an ML Engineer and technical project lead who builds production-grade computer vision systems across edge and cloud. I’ve spent 6+ years taking models from end-to-end development through embedded porting, backend services, and QA—often optimizing for real-time performance and robustness across multiple hardware tiers. At Motive, I lead end-to-end delivery for an in-cabin passenger detection system, including architectural decisions (like moving to full-resolution inference), DRA/discrepancy analysis to close gaps between on-device and offline results, and orchestration of ML/embedded/EFS pipelines. I also enjoy engineering practical tooling that reduces annotation cost and improves quality, such as using VLM-based event gating for human-in-the-loop workflows.

Available to hire

I’m Aman Shekh, an ML Engineer and technical project lead who builds production-grade computer vision systems across edge and cloud. I’ve spent 6+ years taking models from end-to-end development through embedded porting, backend services, and QA—often optimizing for real-time performance and robustness across multiple hardware tiers.

At Motive, I lead end-to-end delivery for an in-cabin passenger detection system, including architectural decisions (like moving to full-resolution inference), DRA/discrepancy analysis to close gaps between on-device and offline results, and orchestration of ML/embedded/EFS pipelines. I also enjoy engineering practical tooling that reduces annotation cost and improves quality, such as using VLM-based event gating for human-in-the-loop workflows.

Skills

Experience Level

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Intermediate

Intermediate

Language

English

Advanced

Hindi

Advanced

Gujarati

Advanced

Work Experience

ML Engineer & Tech Lead (EFS Feature Lead & Technical Project Manager, Passenger Detection Beta) at Motive (Remote, Bengaluru, India)

January 1, 2025 - Present

Served as EFS Feature Lead, Technical Lead, and Technical Project Manager for end-to-end delivery of an AI in-cabin passenger detection system for unauthorized occupant detection. Led multi-task unified driver model development and real-time full-resolution inference (540×960), including firmware porting across device tiers and SNPE-based Qualcomm targets. Drove an architectural switch from cropped to full-resolution inference for full-cabin coverage while mitigating ~10% FPS impact via per-tier model pipelines to preserve SLAs. Improved passenger metric quality from 0.35 to 0.75 mAP. Owned DRA (Detection Rate Analysis) and discrepancy analysis between on-device inference and offline ground truth to identify class confusion/miss-detection gaps and translate findings into a re-training and data-collection plan ahead of GA.

ML Engineer & Tech Lead | AI Feature Lead & Technical Project Manager (Passenger Detection, Beta) at Motive (Remote, Bengaluru)

January 1, 2025 - Present

Serving as EFS Feature Lead, Technical Lead, and Technical Project Manager for an end-to-end AI in-cabin Passenger Detection system for unauthorized occupant detection. Led full-stack delivery including a multi-task unified driver model, uncropped full-resolution inference, firmware porting across device tiers, and QA. Architected a switch from cropped to full-resolution inference to support full-cabin coverage while maintaining real-time SLAs, achieving improved passenger metrics (0.35→0.75 mAP). Owned DRA (Detection Rate Analysis) and discrepancy analysis between on-device outputs and offline ground truth, root-causing class confusion and miss-detection gaps and translating findings into re-training and data collection plans for GA readiness.

ML Engineer & Tech Lead | AI Feature Lead & Technical Project Manager (Passenger Detection) at Motive

January 1, 2025 - Present

Served as EFS Feature Lead, Technical Lead, and Technical Project Manager for end-to-end delivery of an in-cabin Passenger Detection system for unauthorized occupant detection across model development, embedded firmware, backend, EFS pipeline, and QA. Led architecture changes from cropped to full-resolution inference for full-cabin coverage while preserving real-time SLAs by maintaining per-tier model pipelines. Improved passenger detection metric quality from 0.35 to 0.75 mAP and owned detection-rate/discrepancy analysis between on-device inference and offline ground truth to drive targeted re-training and data collection ahead of GA.

ML Engineer & Tech Lead | EFS Feature Lead & Technical Project Manager at Motive

January 1, 2025 - Present

Led end-to-end delivery of an AI in-cabin passenger detection system, covering multi-task unified driver modeling, full-resolution (uncropped) inference at 540×960, embedded firmware porting, backend integration, and QA. Owned architectural decisions to switch from cropped to full-resolution inference for full-cabin coverage while preserving real-time SLAs, mitigating ~10% FPS impact through tiered model pipelines. Drove detection rate analysis (DRA) and discrepancy analysis between on-device inference and offline ground truth to root-cause class confusion/miss-detections and convert findings into a re-training and data-collection plan ahead of GA. Improved passenger metric quality from 0.35 to 0.75 mAP and supported zero-incident style releases at scale.

ML Engineer & Tech Lead / AI Feature Lead & Technical Project Manager at Motive (EFS Feature Lead - Passenger Detection Beta)

January 1, 2025 - Present

Owned end-to-end delivery for an in-cabin Passenger Detection system (unauthorized occupant detection), spanning multi-task model development, embedded/firmware porting, backend, EFS pipelines, and QA. Led a key architecture change from cropped to full-resolution inference for full-cabin coverage while mitigating FPS impact via per-tier model pipelines to preserve SLAs. Improved passenger metric quality from 0.35→0.75 mAP. Built and led DRA (Detection Rate Analysis) and discrepancy analysis between on-device inference and offline ground truth to identify class-confusion and miss-detection gaps, translating findings into a retraining and data-collection plan ahead of GA. Also contributed to hardware portability by deploying SNPE-based Qualcomm-target workflows and maintaining model pipelines across edge device tiers.

Computer Vision Engineer at LG Software (Bengaluru, India)

January 1, 2023 - January 1, 2025

Designed and fine-tuned production computer vision systems, including a multi-task apparel segmentation and tagging solution using SegFormer and ResNet for multi-label classification. Built an automated annotation workflow using SAM combined with CVAT and Roboflow to accelerate training data generation and enable human-in-the-loop review. Achieved strong performance on large-scale fashion datasets (53.2% overall mAP over 25 classes and ~95% accuracy on multi-class tagging). Also developed and deployed real-time edge AI systems on an LG Neural Engine board using YOLOX-tiny for detection and YOLOv8 for instance segmentation, including model conversion pipelines (PyTorch→ONNX→TFLite→LNE) and latency reduction via targeted operator replacements.

Computer Vision Engineer at LG Software

January 1, 2023 - January 1, 2025

Built and deployed production CV pipelines for apparel segmentation and fashion tagging using multi-task deep learning. Designed and fine-tuned a multi-task system leveraging SegFormer with ResNet for large-scale semantic segmentation and multi-label apparel tagging. Implemented an automated annotation pipeline using SAM with CVAT/Roboflow and human-in-the-loop review to accelerate training data generation at scale. Achieved strong performance on multi-class tagging and also worked on edge deployment for LG Neural Engine hardware, covering conversion from PyTorch→ONNX→TFLite→LNE with latency reduction through operator-level optimization (e.g., replacing costly activations/ops).

Computer Vision Engineer at Sysmex (Osaka, Japan)

August 1, 2020 - January 1, 2023

Built and deployed an end-to-end ML pipeline for early-stage Alzheimer’s disease detection using 500+ SMLM images. Applied contrast enhancement, Otsu thresholding, DBSCAN-based protein structure quantification, and engineered geometrical features (>10) for SVM classification, achieving ~97% accuracy. Deployed the optimized solution as a Flask REST API in Docker for scalability. Additionally worked on cancer cell image classification using EfficientNet, handling class imbalance with focal loss/weighting, achieving ~97.2% AUROC and ~95.4% recall. Developed utilities for pose/dataset automation via video analysis using OpenCV, OpenFace, MediaPipe, and a Tkinter GUI to reduce manual preparation time.

Computer Vision Engineer at Sysmex

August 1, 2020 - January 1, 2023

Designed end-to-end ML for early-stage Alzheimer’s disease detection using advanced image processing on 500+ SMLM images. Applied contrast enhancement, Otsu thresholding, DBSCAN-based protein structure quantification, and engineered >10 geometrical features for SVM-based classification, reaching 97% accuracy. Deployed an internal Flask REST API for scalable usage with Docker. Additionally developed an EfficientNet-based cancer cell image classification pipeline addressing class imbalance with focal loss/weighting, achieving strong AUROC and recall, and supported internal research deployment on cancer imaging datasets. Built dataset automation via video analysis using OpenCV/Tkinter plus face recognition (OpenFace) and pose extraction (MediaPipe) to derive pose features for frailty-related indicators.

AI Research Intern at TATA Innovation Labs (Mumbai, India)

May 1, 2019 - August 1, 2019

Developed and trained CNN-based U-Net models for hyper-spectral satellite image semantic segmentation using GIS and geospatial tooling. Achieved ~92% accuracy while reducing training time by about 20 minutes. Optimized feature dimensionality using a genetic algorithm to select top features for mapping key Indian regions, leveraging data processing and storage workflows via QGIS/ERDAS/PostGIS-style pipelines.

AI Research Intern at TATA Innovation Labs

May 1, 2019 - August 1, 2019

Developed deep learning for semantic segmentation of satellite images by training a CNN U-Net model for hyper-spectral imagery. Achieved ~92% accuracy and reduced training time through feature optimization using a genetic algorithm to select top features for mapping key regions. Used GIS tooling and spatial data workflows to support the research pipeline.

Education

M.Tech (Geoinformatics) at IIT Bombay

January 1, 2018 - January 1, 2020

B.Tech (Electrical Engineering) at VGEC – Ahmedabad

January 1, 2014 - January 1, 2018