Overview
Built a high-performance, offline-first educational utility featuring a system-wide floating dot overlay. Integrated a native C++ on-device inference engine using a hybrid local trust model to process local LLM models entirely offline. Implemented real-time optical character recognition (OCR) utilizing Google ML Kit. Designed a multi-threaded, local RAG architecture using a custom TF-IDF keyword-matching algorithm.
Key Features
- ✓Offline-first with on-device C++ LLM inference
- ✓System-wide floating dot overlay
- ✓Real-time OCR using Google ML Kit and MediaProjection API
- ✓Multi-threaded local RAG architecture with sub-10ms latency