Smeet Dhakecha

Research Engineer @ Amazon

Building the future of artificial intelligence through innovative agent frameworks and cutting-edge machine learning solutions. Passionate about transforming complex data into intelligent systems that make a real-world impact.

Explore My Work

Professional Experience

Amazon Web Services - Agentic AI
Software Engineer - Machine Learning
Jan 2024 - Present | Santa Clara, CA
  • Building advanced Agent Simulation frameworks to test and validate AI agent behaviors in complex scenarios
  • Developed LLM-powered systems for real-time sensitive information redaction while maintaining low latency
  • Optimized machine learning models for audio transcript analysis using Large Language Models
  • Maintained scalable transcription pipelines processing large volumes of call data with real-time analytics
Mark and Mary Stevens Institute of Neuroimaging
Graduate Researcher
May 2022 - Jan 2024 | Los Angeles, CA
  • Researched Tau PET imaging for early-onset Alzheimer's disease detection using Partial Volume Correction
  • Developed novel PVC algorithm enhancing disease detection using patient MRI and PET images
  • Leveraged Apache Spark for distributed computing to process PET images in-memory efficiently
  • Implemented parallel processing pipelines using QSUB scripts for large dataset analysis on clusters
ITEMS Institute, USC
Machine Learning Researcher
Sep 2021 - May 2022 | Los Angeles, CA
  • Developed U-Net based models for axon and myelin segmentation from microscopic nerve data
  • Optimized CNN architectures using TensorFlow-Keras through hyperparameter optimization
  • Improved segmentation performance by incorporating Dilated Convolutions
  • Enhanced system latency by 9% through ML solution improvements

Technical Skills

Machine Learning

TensorFlow PyTorch Scikit-Learn Keras LLMs CNNs GANs

Programming

Python Java C++ R Kotlin MATLAB

Data & Analytics

NumPy Pandas Apache Spark Hadoop Tableau Matplotlib

Cloud & DevOps

AWS SageMaker AWS Lambda EC2 S3 AWS Bedrock CloudFormation

Featured Projects

NeurIPS 2024 - BELKA Medicine Prediction
Achieved rank 2/1946 in Kaggle competition predicting protein-protein interaction bindings using encoded molecular structures. Implemented 1D-CNN model with dense layers and dropout optimization.
Python 1D-CNN TensorFlow Kaggle
Retinal Vessel Segmentation for Glaucoma Detection
Developed GAN-based model for blood vessel segmentation from fundus images. Achieved 38% parameter reduction and 4% improvement in quantitative metrics using depth-wise separable convolutions.
GAN CNN TensorFlow Medical AI
ADAS Image De-Hazing System
Implemented Conditional GAN model for efficient haze removal from ADAS images, improving real-time visual detection of pedestrians and cars by 38%.
DCGAN Computer Vision TensorFlow Automotive AI
ASL Fingerspelling Recognition
Ranked 444/1315 in Kaggle competition using Transformer-based model with NLP techniques for 3D keypoint sequence classification in fingerspelled phrase identification.
PyTorch Transformers NLP Computer Vision

Education

Master of Science in Electrical and Computer Engineering
University of Southern California
May 2023

Focus: Machine Learning and Data Science

Bachelor of Technology in Electronics and Communication Engineering
Sardar Vallabhbhai National Institute of Technology, Surat
May 2021

Let's Connect

Interested in collaboration or discussing AI/ML innovations? I'd love to hear from you.

📧 Email 💼 LinkedIn 🔗 GitHub 📅 Schedule a Meeting