Evan Diewald

I am a Data & Machine Learning Engineer at AWS Professional Services, where I help enterprise customers build secure, scalable data infrastructure for media & entertainment, healthcare/life sciences, and manufacturing. I hold an MS from Carnegie Mellon University.

My work spans from academic research in AI for manufacturing to production ML systems. Recent projects include developing deep learning models for the NFL's coverage responsibility system, contributing to AWS open source repositories for genomic language models and embedding fine-tuning, and building AI agents that can autonomously solve software engineering tasks.

I focus on combining industry domain expertise with data science fundamentals for successful AI/ML implementations, with particular interest in GenAI, computer vision, and cloud infrastructure.

Email  /  GitHub  /  Google Scholar  /  LinkedIn

Professional Work

Industry projects and production ML systems.

NFL Coverage Responsibility Model

Deep learning model deployed across all NFL games for real-time coverage analysis. Uses transformer architecture for spatial-temporal dynamics in football plays, enabling coverage assignment identification and matchup analysis. Built in collaboration with AWS and NFL Next Gen Stats.

Serverless GenAI Batch Processing

Serverless architecture for orchestrating large-scale Amazon Bedrock batch inference jobs with 50% cost reduction. Features workflow processing for millions of records using Step Functions, Lambda, and DynamoDB for enterprise AI deployments.

Fine-Tune Caduceus (Mamba-Based DNA Sequence Foundation Model) for Genomic Benchmarks

In this project, we will adapt and fine-tune a pre-trained Caduceus model for DNA sequence classification tasks provided by the Genomic Benchmarks datasets. While these benchmark datasets are publicly available on HuggingFace, we also provide an optional workflow to import the sequences into AWS HealthOmics and pull them into the fine-tuning job. Genomic research organizations who have access to their own, proprietary sequence datasets (e.g. FASTA/FASTQ/BAM format), could use a similar workflow to train custom models. Presented at AWS Analyticon 2025.

Fine-tuning Passage Embeddings with GenQ

Official AWS sample for fine-tuning embedding models using synthetic query generation (GenQ). Features SageMaker integration, Hugging Face Estimators, and production deployment workflows for enterprise ML applications.

Technical Writing

Articles and blog posts on machine learning and data engineering.

Aegis - AI Agent for Solving GitHub Issues

LLM-powered tool achieving 33.67% resolution rate on SWE-Bench Lite, surpassing submissions from Amazon and IBM. Implements advanced agentic workflows with LangGraph for automated code generation and bug fixing in production codebases.

Building a Fantasy Football Research Agent

Complete guide for building AI agents with real-time data integration, Streamlit UI, and AWS CDK deployment. Demonstrates full-stack ML application development with modern agent frameworks and infrastructure-as-code. Update: Check out the Sleeper MCP Server.

Planning the Perfect Hike with NetworkX

Route optimization application using graph algorithms, NASA elevation data, and Dijkstra’s algorithm. Features interactive web application with real-time elevation profiling and mathematical optimization.

Exploring the Helium Network with Graph Theory

Application of Graph Neural Networks using PyTorch Geometric and NetworkX to analyze blockchain data for fraud detection. Implements graph algorithms for identifying suspicious network activity patterns in IoT networks.

LongHive - Smart Beehive Monitoring System

Smart beehive monitoring system using TensorFlow Lite for edge inference. Combines audio classification CNNs, environmental sensors, and Raspberry Pi deployment with Helium protocol integration. Featured in Make: Magazine Vol. 75.

Research

Academic research in computer vision, machine learning, and manufacturing applications.

Spatter Detection and Tracking in High-Speed Video Observations of Laser Powder Bed Fusion

Machine learning approach for automated spatter particle detection using convolutional neural networks in high-speed manufacturing video. Funded by Army Research Laboratory with quantitative insights for process optimization.

Multi-sensor Feature Extraction Framework Using Neural Networks for In-situ Defect Detection in 3D Metal Printing

Patent for innovation in applied neural networks for manufacturing quality control. Demonstrates multi-sensor feature extraction framework for real-time defect detection in additive manufacturing processes.

Additive OS: An Open-Source Platform for Additive Manufacturing Data Management & IP Protection

Open-source platform addressing cybersecurity in additive manufacturing using smart contracts and distributed systems. Features blockchain-based IP protection, peer-to-peer content delivery, NoSQL database with DAG representations, and browser-based GUI with API access.

Deep Learning of Variant Geometry in Layerwise Imaging Profiles for Additive Manufacturing Quality Control

Developed novel deep neural network achieving 92.50% accuracy in real-time flaw detection for 3D metal printing. Features hierarchical dyadic partitioning methodology and shape-to-image registration using CAD files for customized manufacturing quality control.

Design and source code from Jon Barron's website