Projects

Projects

ML/NLP, serverless systems, security, filesystems, and async networking — with measurable performance and reliability wins.

Fake News Detection & Summarization with ML & Transformers

BERTSum • Transformers • scikit-learn • NLTK • PyTorch • ROUGE

NLPTransformersBERTSumSVM/TF-IDFEvaluation
  • Architected an abstractive summarization classifier pipeline using BERTSum, integrating Transformers, scikit-learn, NLTK, & Torch; reduced article length by 63% with >82% semantic retention & a 21% gain in key-point recall.
  • Used TF-IDF + SVM, Naïve Bayes, & Logistic Regression with custom beam search, max length, & dropout tuning to boost ROUGE-Lsum by 18%, achieved an F1-score of 0.86, and delivered a 6% lift over baseline, validated via confusion matrices.

Generative AI-Powered Recipe Web Application

JavaScript • AWS Amplify • Lambda • DynamoDB • Redis • API Gateway • CloudWatch • SQS/SNS

ServerlessGenAIAWSPerformanceObservability
  • Developed a Generative AI web app with JavaScript, AWS Amplify, & serverless Lambda, leveraging DynamoDB and Redis for real-time data processing, achieving sub-200 ms response times for 450+ concurrent queries; optimized API calls with API Gateway, maintaining a 99.9% success rate while supporting 170+ users & ensuring seamless performance through AWS CloudWatch.
  • Decoupled AI inference from API requests via AWS SQS/SNS, reducing API throttling by 12% & bolstering system resilience.

Embedded FAT32 File System Extractor using FUSE (C & Rust)

FUSE • Asynchronous I/O • Thread pools • Rust • Python • Kafka • GraphQL

SystemsC/RustFUSEConcurrencyStreaming
  • Designed a FAT File System Extractor with FUSE in C, parsing 4GB disk images to extract 10,000+ file entries with zero data loss, leveraging asynchronous I/O & optimized thread pools to cut CPU overhead by 21% & boost concurrent file access by 31%.
  • Extended functionality with Rust for memory-safe extensions & Python for cutting extraction verification time by 34%, while integrating Kafka & GraphQL for real-time data streaming & querying, reducing corrupted file errors by 52%.

RESTful Key Management System with JWKS API

Python • Flask • PyCryptodome • OAuth2 • JWT • SQLAlchemy • SQLite

SecurityJWT/JWKSOAuth2FlaskDatabases
  • Devised a RESTful JWKS server in Python using Flask & PyCryptodome to manage 100+ RSA keys & 200+ JWT requests, implementing OAuth 2.0 scopes, access tokens, & key rotation for secure token distribution, achieving a 9% reduction in latency.
  • Crafted a key management system using SQLAlchemy & SQLite for persistent storage of 30+ cryptographic keys, optimizing indexing, eager loading, & connection pooling to improve query throughput by 91% & support secure verification workflows.
  • Revamped server architecture to integrate user authentication using OAuth, processing 50+ registrations to streamline access.

Async Network Server & Proxy System

Python • Asyncio • TLS • TTL caching • Load balancing

NetworkingAsyncioTLSCachingReliability
  • Built an asynchronous HTTPS server in Python, using Asyncio & TLS with 15s TTL caching, reducing duplicate requests by 27% & boosting response efficiency by 5%; fine-tuned load-balancing request routing with target selection, reducing overload by 32%.

RASA — Steampunk Adventure FPS (2D) — Team Lead

Godot • GDScript • Scrum/Agile • A* pathfinding • FSM AI

LeadershipGame DevAgile/ScrumAIGodot
  • Led a 5-member Scrum team using Agile methodologies to design & ship RASA — a 2D Metroidvania-style FPS game in Godot, showcased at UNT Demo Day to 500+ attendees, delivering immersive gameplay & sophisticated enemy interactions while sustaining 60 FPS during fight scenes.
  • Used GDScript to integrate A* pathfinding algorithm with hierarchical FSMs to improve NPC path accuracy by 19%.