NoteForge AI

🧠 Autonomous YouTube Knowledge Synthesis Architecture

NoteForge AI addresses the high cost and contextual latency of processing long video transcripts by implementing an Enterprise Hybrid Multi-LLM Router. High-complexity tasks—such as single-shot synthesis of flashcards, deep chapter study guides, and comprehensive quizzes—are routed to Gemini 2.5 Flash. Lightweight structural tasks—such as Table of Contents extraction and mind map branch reorganization—are delegated locally to Ollama (qwen2.5-coder), minimizing cloud API costs while maintaining instant responsiveness.

⏱️ Deterministic Transcript Slicing vs. Vector Overheads

Traditional RAG architectures rely heavily on expensive external vector databases (such as Pinecone or Weaviate) to retrieve relevant chapter context, introducing latency and risk of chunk boundary misalignment. NoteForge AI solves this through Deterministic Timestamp Slicing. By extracting native YouTube chapter timestamps and aligning transcript boundaries mathematically, the system isolates exact transcript segments with zero millisecond latency, zero computational overhead, and guaranteed contextual accuracy without hallucinations.

🛡️ Enterprise Hot-Swap Key Rotation & State Persistence

To handle unpredictable API rate limits during batch playlist processing, NoteForge AI features a Round-Robin Multi-Key Pool with automated exponential backoff powered by Bottleneck. If Key A encounters a 429 RESOURCE_EXHAUSTED error, the engine marks the key with an exhaustedUntil timestamp and seamlessly hot-swaps to Key B without interrupting active generation. Furthermore, job state and chapter progress are persisted locally via Prisma and SQLite, allowing interrupted jobs to resume instantly at the exact missing chapter.

🎨 Dynamic Headless PDF & Mind Map Render Engine

NoteForge AI provides print-ready document exports through an integrated Puppeteer headless rendering pipeline. The system dynamically injects customizable runtime CSS themes (Modern Bento, Academic Serif, Corporate Blue, and High-Contrast Dark Mode) with strict page-break-inside: avoid layout guards for inline code blocks and Mermaid diagrams.