CS @ UW-Madison (Dec 2025) | Open to Work

Portfolio Website 📧 yajatnarayan@gmail.com
🔗 LinkedIn
📍 Madison, WI

Developing Unity-based VR training software for U.S. Air Force personnel as part of a $15M defense contract

• Built real-time 3D spatial replay system using Apple Vision Pro sensor data with sub-centimeter positional accuracy
• Engineered Euclidean transformation algorithms on custom Unity mesh models for training evaluation
• Reduced design iteration cycles by 40% through user research with 15+ Air Force personnel and rapid prototyping

Architected cloud infrastructure and CI/CD pipelines for IoT data processing

• Built decoupled CI/CD pipeline processing 100K+ daily data points using GoLang, AWS Lambda, and DynamoDB
• Increased data processing throughput by 35% with asynchronous workers and optimized batch operations
• Reduced production downtime by 45% through automated validation and health monitoring systems

Languages: Python, GoLang, JavaScript/TypeScript, Swift, Java, C
Frontend: React, Unity, SwiftUI, HTML5 Canvas
Backend & Cloud: Node.js, Express, AWS (Lambda, DynamoDB, S3), Docker, Flask
Security: AES-256-GCM, scrypt/Argon2, OWASP standards, rate limiting, HMAC verification
Tools: Git, CI/CD pipelines, PyTorch, NumPy, Chrome Extension API

Production-grade password vault with military-grade encryption, browser autofill, and zero-plaintext storage

Tech: TypeScript, Node.js, React, Chrome Extension API, AES-256-GCM
Security Features:

• OWASP-compliant encryption with scrypt KDF (65,536 iterations) and HMAC-SHA256 integrity verification
• Rate limiting (5 attempts/15min), auto-lock sessions, cryptographically secure password generation
• Atomic file operations with zero-plaintext storage—vault data only decrypts in memory during active sessions
• Comprehensive test coverage for encryption, validation, rate limiting, and domain matching

Impact: Built as an educational MVP demonstrating secure password management principles inspired by 1Password

Full-stack OCR application with custom neural network—no ML frameworks

Tech: Python, NumPy, Flask, JavaScript, HTML5 Canvas
Implementation:

• Multi-layer perceptron built from first principles using only NumPy (no TensorFlow/PyTorch)
• Custom backpropagation, gradient descent, and activation functions demonstrating deep understanding of neural network mathematics
• Interactive web interface with real-time digit drawing, preprocessing user input into 400-dimensional feature vectors
• Flask RESTful API supporting dual-mode functionality: model training with labeled samples and real-time prediction

Impact: Demonstrates ML fundamentals and full-stack integration without relying on high-level frameworks

I focus on building systems that solve real problems with measurable impact. Whether it's reducing military training risk through VR simulations, securing user credentials with production-grade cryptography, or optimizing data pipelines to handle 100K+ daily operations, I prioritize:

• Security-first architecture — OWASP compliance, threat modeling, defense-in-depth
• Measurable outcomes — Quantified improvements in efficiency, reliability, and user experience
• Production quality — Comprehensive testing, atomic operations, proper error handling

Currently exploring: Spatial computing, real-time distributed systems, and infrastructure automation.

Open to full-time Software Engineering roles starting January 2026 | Interested in backend systems, cloud infrastructure, security engineering, and VR/AR development