Voice-Controlled Robot
Edge-deployed AI robot on Raspberry Pi 5 with Picovoice wake-word (<10ms), OpenAI Whisper ASR, and Gemma-3-1B intent parsing. MediaPipe + OpenCV human-following at 5 FPS via ROS 2 Jazzy.
Edge-deployed AI robot on Raspberry Pi 5 with Picovoice wake-word (<10ms), OpenAI Whisper ASR, and Gemma-3-1B intent parsing. MediaPipe + OpenCV human-following at 5 FPS via ROS 2 Jazzy.
RAG pipeline using llama3.2:3b with Selenium crawler scraping 10K+ JS-rendered blocks. Chunked and embedded via Sentence Transformers, indexed with FAISS for sub-1s retrieval.
Real-time C++ OpenCV pipeline detecting lanes (RANSAC, Kalman Filter), vehicles (Haar Cascades), traffic signs & pedestrians (YOLO). Sub-40ms latency for high-speed driving.
Twitter's recommendation algorithm simulation on synthetic data. Bipartite graph construction, SALSA-style random walks, and feature-based tweet scoring with author bonuses.
Dockerized Django app on GCP with PostGIS mapping 150+ composting sites via OSM Overpass API. React+Bootstrap UI with authentication and impact tracking leaderboard.
Production-ready AI system with pluggable storage (SQLite/MongoDB Atlas). Features drift decisioning, novelty filtering, and budget-aware assembly for optimal retrieval.
Desktop health app built at CalHacks 11.0. Real-time slouch detection and emotional expression analysis using MediaPipe and Hume AI for personal wellness monitoring.
Coming soon — technical deep-dives and lessons from the field.