Neuronote is an AI-powered knowledge structuring engine that converts unstructured text into typed entities (tasks, insights, people, temporal references) for structured retrieval and downstream reasoning. I designed a hybrid retrieval system combining semantic embeddings with lexical boosting and weighted scoring, later replacing external vector infrastructure with an in-app similarity pipeline to reduce infrastructure cost by ~50% while maintaining accuracy. The system includes dynamic similarity thresholding and reranking to balance recall against hallucination risk, along with a context-aware summarization pipeline via Gemini. Client–server synchronization was optimized through local caching and mutation reconciliation, reducing server load by ~40% while preserving data consistency.