Transform Enterprise Knowledge with RAG + LLM Intelligence
Document Master bridges your proprietary corporate knowledge repositories with state-of-the-art Large Language Models. Query manuals, procedures, APIs, and tickets with instant answers backed by exact source citations.
🧠 5 Core LLM & RAG Integration Points
1. Map-Reduce Document Summarization
During document indexing, long text is broken into optimal chunks. The LLM generates concise summaries for each chunk, which are hierarchically combined using Map-Reduce into a comprehensive document overview stored directly in your partition index.
2. Query Auto-Correction & Optimization
Never miss a result due to typos or phrasing mismatches. If an initial search yields zero matching document references, our LLM auto-correction prompt rephrases the query to resolve misspelled terms (e.g. "Pintung" to "Pingtung") and re-runs the retrieval automatically.
3. Conversational Cited RAG Q&A
The workspace chat console feeds retrieved document context snippets alongside the user prompt to the LLM, synthesizing clear, accurate answers strictly backed by explicit document filename citations (e.g. [policy_2026.pdf]).
4. Conversation Memory Summarization
As multi-turn Q&A threads grow long, the backend automatically summarizes past turns into a compact context memory snippet, keeping the LLM prompt well within token memory limits while preserving complete chat history.
5. Hybrid Vector & Phonetic Search
Combines high-dimensional vector embeddings (cosine similarity) with term-frequency keyword density, Metaphone phonetic matching, and Levenshtein fuzzy querying for resilient reference extraction.
🛡️ Enterprise Architecture & Multi-Source Connectors
🔒 Zero-Local-Footprint Security
No compiled document indexes or summaries are stored on our servers. All partition files are encrypted and stored directly in your own configured Google Drive, Dropbox, or FTP storage.
🔗 Multi-Source Knowledge Connectors
Connect repositories seamlessly from Google Drive, Dropbox, FTP/SFTP servers, Confluence Spaces, and Jira Projects via OAuth 2.0.
📄 Native Multi-Format Parsers
Parse PDFs, Microsoft Word (.docx), LibreOffice Writer (.odt), eReader Ebooks (.epub), Markdown, HTML, CSV, and plain text natively using secure C-compiled standard libraries.
⚡ Global Workspace Index Map
Scales to 3,000+ files effortlessly by using a lightweight reverse word map (doc_0_index.json) to filter queries down to relevant files in milliseconds.