Machine Learning Engineering Intern

Usman Khan

usmankhan.dev *****@*********.*** US Citizen linkedin.com/in/khanu github.com/ukhan1219 Education

University of Central Florida Orlando, Florida

B.S. in Computer Science 3.8/4.0 GPA Expected Graduation: December 2025 Relevant Coursework: Algorithms in Machine Learning, Artificial Intelligence/Machine Learning, Robot Vision, Computer Vision Technical Skills

Languages: Python, Rust, TypeScript/JavaScript, C, Java, SQL, NoSQL Frameworks: PyTorch, Keras, TensorFlow, NumPy, Pandas, SKLearn, Next.js, Node.js, Express.js, React, Tailwind Tools: Git, Github, Docker, Linux, LaTeX, Prisma, Neo4J, Figma, Amazon Web Services, Google Cloud Platform Work Experience

Software Engineering Intern Aug 2024 – Jun 2025

Vcom3D — Python, TensorFlow, OpenCV, Raspberry Pi 5, Meta Quest 3, BioGears (UW), C++, XML Orlando, Florida

• Built pose tracking models using TensorFlow on Raspberry Pi, boosting accuracy & reducing latency by 30%

• Merged BioGears (University of Washington) for injury simulation, boosting training realism by 98% across modules

• Created AR/VR apps on Meta Quest to support simulations ran by BioGears in a distributed system architecture

• Refined system integration across components via cross-functional collaboration, slashing errors & streamlining updates Machine Learning/AI Undergraduate Research Assistant Apr 2024 – Apr 2025 University of Central Florida — Python, TensorFlow, Neo4J, NumPy, SKLearn, NetworkX, Pandas Orlando, Florida

• Enforced automated distributed data mining algorithms using AI/ML via Neo4J for enhanced predictive analytics

• Generated data mining methods for RandomForestRegressor on a DARPA dataset (6.8M+ nodes) to detect illicit activity

• Devised scalable distributed data pipelines boosting entity tracking accuracy and speed by 30% across datasets

• Deployed statistical methods for performance optimization, reducing processing time by 40% for high-volume pipelines Projects

PyChess Python, PyTorch, Hugging Face Transformers/TRL, Accelerate, python-chess, Stockfish, TensorBoard, DistilGPT-2

• Engineered an end-to-end chess AI post-training pipeline that automates data generation and training with robust tracking

• Processed 90M positions; curated 1M supervised samples and 500k preference pairs using strict quality filters

• Reduced data generation time by 97% via multithreading and optimized I/O; eliminated memory-related training failures Mantle SwiftUI, Python, PyTorch, Core ML, Transformers, Hugging Face, Metal (MPS), Amazon Web Services EC2

• Converted Transformer models (Mistral, Llama) from PyTorch to Core ML utilizing AWS EC2 instances

• Applied Core ML compression (quantization, pruning, palettization) shrinking models by 75% while retaining accuracy

• Accelerated inference 25% leveraging Metal Performance Shaders (MPS) optimization on for On-Device inference

• Developed privacy-first SwiftUI app (iOS 18+) for On-Device ML inference, enabling offline AI chatbot functionality Glance SwiftUI, Golang, Firestore, Firebase Auth, Plaid API, Google Cloud Platform, Figma, XCTest

• Architected a budgeting app using SwiftUI and a Go backend, achieving seamless Plaid API integration

• Implemented secure authentication via Firebase Auth & managed sessions, supporting 100+ concurrent users reliably

• Enhanced data retrieval speeds by 40% through strategic caching & optimized Firestore queries in the Go backend DUI Rust, Crates.io (Cargo), Homebrew, clap, rustyline, crossterm, tui-rs, serde

• Reimagined a Docker CLI with 100% command parity, interactive mode, and real-time charts/dashboard

• Published on Cargo as dui-cli and Homebrew tap; reached 150+ downloads

• Improved UX with tab completion, contextual help, smart suggestions; optimized release builds Whooga PERN: PostgreSQL, pgvector, Express.js, React, Node.js, Python, TypeScript, AWS RDS

• Managed full stack project with 4-person team building specialized collections marketplace using Jira

• Architected vector database with pgvector for semantic search; engineered ML pipeline for item matching with BERT

• Designed computer vision system (YOLOv8, SAM, DINOv2) for semantic image matching across photography conditions

• Implemented real-time messaging system with WebSockets enabling seamless buyer-seller communication for transactions

Contact this candidate