Basmo Chatbook

Ingests book text (via manual upload, OCR, or ISBN lookup) and creates a searchable, semantically-indexed knowledge base that enables the AI to retrieve relevant passages during conversation. The system likely uses vector embeddings (sentence or paragraph-level) to map book content into a high-dimensional space, allowing retrieval-augmented generation (RAG) to ground responses in actual book text rather than relying solely on the model's training data. This prevents hallucination by anchoring answers to source material.

Maintains a multi-turn conversation context while dynamically retrieving relevant book passages to answer user questions. The system uses a context window (likely 4K-8K tokens) to track conversation history, combines it with real-time semantic search over the indexed book, and generates responses that cite specific passages. This prevents the chatbot from drifting into general knowledge and ensures answers remain grounded in the book's actual content, reducing hallucination risk compared to vanilla LLM chat.

Accepts books in multiple formats (PDF, EPUB, image scans, ISBN lookup) and automatically converts them into machine-readable text using OCR (optical character recognition) for scanned books or native text extraction for digital formats. The system likely uses a cloud-based OCR service (e.g., Tesseract, AWS Textract, or proprietary) to handle low-quality scans, with fallback logic to retry failed pages or prompt users to re-upload clearer images. This enables users to add physical books to their library without manual transcription.

Maintains a user's personal library of indexed books with metadata (title, author, ISBN, cover image, reading progress, tags, notes) and enables browsing, searching, and organizing books by category, rating, or custom collections. The system likely stores metadata in a relational database (user → books → chapters/sections) and provides a UI for library management. This allows users to manage multiple books and switch between them in conversations without re-uploading.

Enables users to search for concepts, themes, or passages across an indexed book using natural language queries rather than keyword matching. The system converts the user's query into a vector embedding and performs similarity search against the book's indexed passages, returning the most relevant sections ranked by semantic relevance. This allows users to find discussions of a topic even if they don't know the exact wording used in the book.

Automatically generates summaries of books or chapters and extracts key insights, themes, and arguments using the LLM. The system likely uses the indexed book content as context, prompts the LLM to identify main ideas and supporting evidence, and presents summaries at multiple granularities (full book, chapter, section). This allows users to quickly grasp a book's core ideas without reading the entire text.

Supports extended conversations where users ask follow-up questions, request clarifications, and explore ideas in depth. The system maintains conversation history, tracks which passages were cited in previous responses, and allows users to ask the AI to re-examine or reinterpret passages based on new context. This enables Socratic-style learning where users progressively deepen their understanding through dialogue.

Reduces AI hallucination by requiring the LLM to cite specific passages from the indexed book when answering questions. The system uses a retrieval-augmented generation (RAG) approach where the LLM is prompted to only answer based on retrieved passages and to explicitly state when information is not found in the book. This creates accountability and allows users to verify answers against source material.

Tracks which books a user has read, is currently reading, or wants to read, and optionally provides personalized recommendations based on reading history and preferences. The system stores reading progress (percentage complete, last read date, bookmarks) and may use collaborative filtering or content-based recommendation algorithms to suggest similar books. This helps users discover new books and maintain reading momentum.

Allows users to export passages, summaries, or conversation transcripts in standard citation formats (APA, MLA, Chicago) for use in essays, research papers, or presentations. The system likely integrates with citation management tools (Zotero, Mendeley) or generates citations programmatically based on book metadata. This enables seamless integration with academic workflows.