Docoline
A high-performance document management API with redlining, search, versioning, and AI-assisted legal tools.
Overview
Docoline is a proof-of-concept document editing and redlining service designed for legal workflows. Users can add new documents, highlight specified text for editing or redlining, and track changes through document history. Documents are scoped per session, so edits are discarded when the session ends.
AI Features
- Document Q&A with citations: ask questions about a document and get answers grounded in cited passages. Retrieval chunks documents on paragraph boundaries and embeds them with Gemini's embedding model.
- AI-generated summaries: concise summaries of full documents on demand.
- AI text suggestions: rewrite suggestions for highlighted text, guided by the user's prompt.
- Fast inference: AI inference runs on Groq's LPUs (Language Processing Units, purpose-built for text workloads) alongside Gemini LLMs.
- PII scrubber: personal information such as SSNs, emails, and phone numbers is removed before any text is sent to Gemini's AI models.
- AI evals: built-in evaluations for the PII scrubber, retrieval quality (recall@k and MRR), and answer faithfulness.
Search
- Semantic search: meaning-based matching powered by embeddings.
- Exact-word search: keyword matching backed by an inverted index.
- Flexible scope: search across all documents or scoped to a single document.
Editing & Redlining
- Redlining: highlight specified text for editing with tracked changes and document history.
- Unified editing: multiple changes are bulked into a single edit and applied with a right-to-left strategy, preventing offset drift across multi-change operations.
- Versioning: document history tracking across edits.