Looking for a job

Aman Kumar Sharma

@amankumarsharma1

AI research engineer at EdgeRunner AI who brought on-device multimodal models to Army Research Laboratory robots.

At EdgeRunner AI, I deploy multimodal vision-language models on Clearpath Warthog and Ghost Robotics Vision 60 robots in collaboration with the US Army Research Laboratory. I trained models to interpret their surroundings and follow plain-language commands through ROS 2.

I also built evaluation and data pipelines for EdgeRunner models and trained Tactical models with SFT, LoRA, distillation, and reinforcement learning. EdgeRunner 20B reached GPT-5 parity on military tasks while running on the edge.

Earlier, at Vision Labs’ Autonomous Vehicle Project (ALIVE), I developed camera–LiDAR perception for an on-road autonomous vehicle. I optimized and quantized models for real-time 30 FPS performance on NVIDIA AGX, and worked on transferring perception from simulation to road conditions.

My earlier research and engineering work included efficient Transformers, computer vision, and robotics. I co-authored four arXiv papers on edge LLMs.

Experience

Work history, roles, and key accomplishments

Current

AI Research Engineer

Current

EdgeRunner AI

Jul 2024 - Present (2 years 4 months)

Deployed multimodal vision-language models on-device on ground robots and UAVs for defense applications. Developed data synthesis pipelines, evaluation benchmarks, and training methods (SFT, LoRA, RL) enabling small edge models to match larger ones.

ROS2 PyTorch Jax TensorFlow SFT LoRAs Reinforcement Learning DPO PPO GRPO Policy Distillation Data Synthesis Evaluation Pipelines VLMs Multimodal Transformers AGX Jetson Docker Linux

Independent Researcher, Efficient ML

University of Maryland

Aug 2023 - May 2024 (9 months)

Reduced cost of base Transformer model for multimodal LLMs by 50% with hardware-friendly variant. Applied efficient-model techniques including block sparsity and token pruning, achieving 2x speedup with lottery-ticket sub-network.

Transformers Structured Sparsity Token Reduction Lottery Ticket Hypothesis EfficientML Multi Modal LLMs

Teaching Assistant, Reinforcement Learning and Computer Vision

University of Maryland

Jan 2023 - May 2024 (1 year 4 months)

Designed assignments and projects on AI, deep learning, reinforcement learning, ROS, and computer vision. Delivered course content and evaluated student work.

Reinforcement Learning Computer Vision ROS Open AI Gym Deep Learning Image Processing Vision Language Models

Machine Learning/Computer Vision Intern

VecTech Inc.

May 2023 - Aug 2023 (3 months)

Improved F1 score by 9.1% on bio-medical data with hybrid ensemble. Built semi-automatic segmentation labeling pipeline reducing annotation time by 90%.

Machine Learning Computer Vision Ensemble Learning Segmentation Model Interpretability

Graduate Research Assistant, Computer Vision

University of Maryland

Oct 2022 - Mar 2023 (5 months)

Achieved 80% AP (+48%) on instance segmentation of small RGB-D dataset by benchmarking Transformer models in Detectron2. Accelerated segmentation labeling with semi-automated annotation pipeline using eye tracking in AR/VR.

Computer Vision Instance segmentation Detectron2 Facebook DETR MaskFormer Mask2Former AR VR Eye Tracking

Perception Research Engineer

IIIT Delhi

Dec 2020 - Jun 2022 (1 year 6 months)

Developed and deployed camera + LiDAR perception module for multi-obstacle detection and tracking on autonomous vehicle. Optimized models with TorchScript and TensorRT achieving real-time 30 FPS at 41% mAP on NVIDIA AGX.

Camera LiDAR Perception Object Detection Trackings TorchScript TensorRT AGX ROS Domain Adaptation Self Training Grad CAM Generative Models

Artificial Intelligence Research Engineer

Secure Meters Pvt. Ltd.

Jan 2020 - Dec 2020 (11 months)

Deployed Transformer-based PPE detection system and real-time geo-fenced hazard warning system on Jetson Nano. Reduced quality-control cycle time by 12% with anomaly detection.

Facebook DETR YOLOv3 NVIDIA Jetson Anomaly Detection Isolation Forest Computer Vision

Deep Learning/Machine Learning Engineer (Part-Time)

IIIT Delhi

May 2020 - Aug 2020 (3 months)

Prototyped night-time driver drowsiness detection system on NIR camera with custom lightweight CNN optimized for Jetson Nano. Led team building Android app for driver-distraction prevention using ML Kit.

Deep Learning CNN NVIDIA Jetson Android ML Kit Computer Vision

Education

Degrees, certifications, and relevant coursework

University of Maryland, College Park

Masters in Robotics, Robotics

2022 - 2024

Grade: 3.66/4

Masters in Robotics from the University of Maryland, College Park, with a CGPA of 3.66/4. Relevant coursework includes ML for Cybersecurity using LLMs, Deep Learning, Soft Dev for Robotics, and Computer Vision.