Dylan Geraci

Architecting a FastAPI backend with 30+ REST endpoints, real-time WebSocket messaging with JWT auth, and MongoDB persistence for an AI-powered task marketplace. Integrated Google Gemini AI for NLP-based query parsing to improve task matching accuracy.

Contributed to a real-time smart cooking assistant for low-income supportive housing — a multi-semester research project deploying a fine-tuned YOLOv11s model on a Raspberry Pi 5 with an AI HAT+ NPU for edge-based cooking state detection. The system uses a custom dataset of 1,300+ labeled images, temporal majority-vote stabilization, and async API integration to guide residents step-by-step through recipes via computer vision.

Rebuilt a multi-threaded Python system orchestrating 16 microcontrollers for real-time AI interaction, and engineered a speech recognition pipeline using whisper.cpp achieving 2–4s response latency. Deployed a production AI installation serving continuous live interaction during a multi-month museum exhibition.

Mentoring 40+ STEM transfer students through 1-on-1 advising, technical guidance, and academic planning. Delivering technical presentations and workshops to 100+ students on software engineering and career development.

Serving as a board member for the AI Team, conducting Data Science and ML workshops for 35+ students covering feature engineering, model training, and evaluation. Increased workshop attendance by 50% by developing real-world programming demo applications.

Taught MIT AI concepts using Scratch, DALL·E, and ChatGPT, guiding students through hands-on activities and prompt-based AI interactions in a collaborative learning environment.

Led a team of 10 while providing daily support to 800+ members, consistently exceeding 130% of monthly revenue goals.