>DeepNote
AI-powered study tool to interact with PDFs, generate answer keys from exam papers, chat with notes, and auto-create flashcards; backed by Flask APIs and Pinecone for embedding storage, with Supabase buckets for file storage.
// specifications
Technical Architecture Matrix
// blueprint
Detailed Engineering Breakdown
DeepNote is an intelligent synthesis workspace designed to transform static study materials—textbooks, lecture transcripts, and past exam papers—into active conversational environments. Rather than passive reading, students can test their comprehension, generate structured study guides, and extract verified answer keys directly from source texts.
//Architecture & Pipeline:
▸Document Ingestion & Chunking: PDFs uploaded through the Next.js interface are securely stored in Supabase Storage buckets. A Flask microservice extracts document text, segments passages into semantic paragraphs, and generates dense vector embeddings using Google Gemini embedding models.
▸Vector Indexing & Retrieval: Embeddings are indexed in Pinecone with metadata tagging by document, page number, and section title. Retrieval-Augmented Generation (RAG) utilizes hybrid search to match conceptual queries against specific textbook citations.
▸Synthesis Engine:
▸ - Exam Paper Key Generation: The model parses exam prompts, searches relevant syllabus passages, and drafts step-by-step solutions with direct reference citations.
▸ - Active Recall Generation: Automatic flashcard generation identifies core definitions, formulas, and contrasting concepts, formatted for spaced repetition review.
▸ - Contextual Note Chat: Grounded Q&A ensures conversational answers strictly reflect the uploaded course materials.
Architecture Highlights:
▸Streaming Responses: Real-time response streaming ensures immediate interactivity during long-form answer generation.
▸Strict Grounding: Prompt engineering constraints minimize hallucinations by enforcing inline page and section citations for every synthesized explanation.
// stack