BUILDING AI SYSTEMS AT SCALE

AMEYA
DHAMANASKAR

Deploying AI from Research to Production →

ML Engineer at Optum, building AI-driven systems to optimize pharmacy operations using forecasting, simulation, and reinforcement learning. Previously developed and deployed production-grade computer vision models across 2,000+ edge devices.

SEE MY WORK → GET IN TOUCH
2K+
Devices Deployed
>98%
Model Accuracy
40%
Faster Inference
25%
↑ Throughput
// 01_ABOUT

WHO I AM

Ameya Dhamanaskar - ML Engineer

Quick Facts

  • Based in Phoenix, AZ
  • MS CS from ASU (4.0 GPA)
  • Published in Pattern Recognition
  • 5+ years building ML systems
  • PyTorch • TensorRT • Docker

I'm an ML Engineer specializing in taking AI systems from research to real-world production. My work sits at the intersection of computer vision, edge deployment, forecasting, and reinforcement learning—with a focus on building systems that perform reliably at scale.

I previously conducted computer vision research at the Institut de Robòtica in Barcelona, publishing in Pattern Recognition and working on 3D pose estimation and self-supervised learning. Since then, I've shifted from research to impact—shipping production ML systems, optimizing inference, and deploying on edge hardware.

At Optum, I've built computer vision systems for pharmacy automation (YOLOv8, TensorRT, DeepSORT), deployed models to 2,000+ NVIDIA Jetson devices, and now lead efforts to optimize pharmacy throughput using forecasting, discrete-event simulation, and RL.

I care about problems where ML meets real-world constraints—latency, memory, compute, reliability—and where combining multiple techniques (vision + forecasting + RL) unlocks meaningful operational improvement.

Currently
Designing AI-driven systems to optimize end-to-end order flow across pharmacy production lines

// 02_EXPERIENCE

WHERE I'VE BUILT

ML Engineer
Optum (UnitedHealth Group)
2023 - Present
2,000+
DEVICES
25%
↑ THROUGHPUT
40%
FASTER

Architected AI optimization for pharmacy automation. Deployed YOLOv8 + TensorRT on NVIDIA Jetson edge devices.

• Designed scalable CV pipelines combining synthetic and real-world datasets, maintaining >98% accuracy

• Optimized with TensorRT FP16/INT8 quantization, increasing inference 40% and reducing memory 30%

• Containerized inference with Docker for reproducible deployment

• Led design reviews and mentored 2 junior engineers

PyTorch YOLOv8 TensorRT DeepSORT Docker NVIDIA Jetson
ML Intern
Radius AI
2022 - 2023
+10
mAP GAIN
50%
↓ LATENCY
80%
↓ ANNOTATION

Enhanced retail product detection using 3D sim2real transfer learning with RetinaNet and Faster R-CNN.

• Achieved +10 mAP improvement while halving inference latency

• Optimized Region Proposal Networks via pruning and mixed-precision inference

• Leveraged SAM for automated labeling, doubling ViT training efficiency

RetinaNet Faster R-CNN SAM ViT
ML Researcher
Institut de Robotica (CSIC-UPC) • Barcelona
2019 - 2021
+12%
SOTA ACCURACY
150K
FRAME DATASET
2023
PUBLISHED

Self-supervised egocentric 3D pose estimation combining first- and third-person views. Pattern Recognition, 2023.

3D Pose Estimation Self-Supervised Learning Published Research
// 03_PROJECTS

SELECTED WORK

[YOLOv8 Detection GIF]
Real-time pharmacy automation

Pharmacy Automation CV System

Production computer vision pipeline deployed on 2,000+ NVIDIA Jetson devices. Optimized YOLOv8 with TensorRT achieving 40% faster inference while maintaining 98%+ accuracy.

YOLOv8 TensorRT Edge ML PyTorch
[3D Pose Visualization]
Egocentric human pose

Egocentric 3D Pose Estimation

Self-supervised framework for 3D human pose estimation from egocentric viewpoints. Combines first- and third-person cues with temporal CNNs. Published in Pattern Recognition 2023.

3D Vision Self-Supervised PyTorch Research
[Retail Detection Output]
Product recognition

Sim2Real Product Detection

Enhanced retail product detection using 3D sim2real transfer learning. Achieved +10 mAP with RetinaNet while reducing latency by 50%. Integrated SAM for 80% faster annotation.

RetinaNet SAM Sim2Real ViT
// 04_TECHNICAL_SKILLS

WHAT I USE

ML & Computer Vision

  • Object Detection & Tracking
  • 3D Pose Estimation
  • Segmentation & Classification
  • Vision Transformers
  • Sim2Real Transfer Learning
  • Self-Supervised Learning

Frameworks & Tools

  • PyTorch (Primary)
  • TensorRT (Optimization)
  • ONNX (Interop)
  • OpenCV (Vision)
  • CUDA (GPU Programming)
  • Docker (Deployment)

Production & MLOps

  • Model Quantization (FP16/INT8)
  • Pruning & Compression
  • CI/CD Pipelines
  • Model Registry & Versioning
  • Monitoring & Observability
  • Edge Deployment (Jetson)
// 05_EDUCATION

ACADEMIC BACKGROUND

ASU
LOGO

Master of Science, Computer Science

Arizona State University
GPA: 4.0/4.0 • Focus: Machine Learning, Computer Vision, Deep Learning
2021 - 2023

Bachelor of Engineering, Electrical and Electronics

Birla Institute of Technology and Science (BITS Pilani)
Focus: Data Structures and Algorithms, Operating Systems, Object-Oriented Programming
2014 - 2018