Document Intelligence

Roovie can parse uploaded documents and turn them into knowledge that both users and AI agents can query.

Upload a PDF, Word document, or text file. Roovie sends it through an AI parsing pipeline that extracts the full content and structure. Once parsed, the document becomes conversational — users can ask questions in natural language and get answers grounded in the actual document text, with citations pointing to specific pages.

This is not a search engine that returns keyword matches. It is a comprehension layer that reads the document, understands the question, and formulates an answer using the document's own content as evidence.

In One Line

Upload a document. Parse it with AI. Ask questions. Get answers with page-level citations.

How It Works

Upload a document (PDF, Word, text, markdown, EPUB)
  → AI parsing extracts structured content
  → Document becomes queryable
  → Users ask questions in natural language
  → AI answers with citations to specific pages
  → Parsed content available as structured JSON for agent consumption

Supported File Types

The system supports six document formats:

• PDF (.pdf) — the most common format for specifications, reports, and technical documents
• Microsoft Word (.doc, .docx) — project documentation, proposals, and narratives
• Plain text (.txt) — simple text files and notes
• Markdown (.md) — technical documentation and structured notes
• EPUB (.epub) — digital publications and standards documents

The Parsing Pipeline

When a document is uploaded, parsing can be triggered manually with a single click or automatically during upload.

The pipeline progresses through four stages:

1. Pending — the document is queued for processing
2. Uploading — the file is being transmitted to the parsing service
3. Processing — AI is analyzing the document structure, extracting text, identifying sections, and building a queryable representation
4. Parsed — the document is ready for questions

The interface polls automatically during processing, showing real-time status updates. If parsing fails, an error message explains why, and a re-parse button lets the user try again.

Once parsed, the system reports the page count and makes the document available for conversation.

Asking Questions

After parsing, users interact with the document through a chat interface.

The Conversation

Type a question in natural language. The AI reads the parsed document content, finds the relevant sections, and formulates an answer. Each answer can include:

• Formatted text — with headers, lists, bold, italic, and code blocks rendered properly
• Citations — expandable references to specific pages with excerpt text showing the exact passage the answer is based on

Citations are numbered and collapsible. Each citation shows the page number and a text excerpt, so users can verify the answer against the source material.

Suggested Questions

When opening a document chat for the first time, the system generates suggested questions based on the document's content. These serve as starting points — click one to populate the input field and send it.

Suggestions are generated per document, so a mechanical specification will suggest different questions than a commissioning report or an energy audit.

Chat History

Conversation history is preserved per document. Returning to a previously queried document shows the full conversation with all questions, answers, and citations intact.

The history can be cleared to start a fresh conversation.

The Document Workspace

For deeper analysis, the Document Chat Workspace provides a three-panel layout:

• Left panel — a list of all parsed documents, filtered to show only those ready for Q&A. Click to select.
• Center panel — a document viewer showing the original file content for visual reference while chatting.
• Right panel — the chat interface for asking questions about the selected document.

This layout allows users to read the document and ask questions simultaneously, without switching between views.

The workspace supports switching between Suggested and Custom question modes — suggested questions provide AI-generated starting points, while custom mode gives an open text input for freeform queries.

Structured Output for Agents

Parsed document content is not only available through the chat interface. It can also be downloaded as structured JSON.

This is significant for two reasons:

Agent Consumption

When Roovie's AI design agents encounter a question they cannot answer from the building model alone — such as "what does the specification say about minimum insulation R-value?" or "what are the commissioning requirements from the project manual?" — they can reference parsed document content to inform their responses.

The structured JSON format makes document knowledge machine-readable, not just human-readable. Agents can consume it programmatically to ground their design decisions in project-specific documentation.

Portable Knowledge

The JSON export means the parsed content is not locked inside Roovie's interface. It can be used in external tools, reports, or integration workflows where structured document data is needed.

How Parsing Works Under the Hood

The parsing service accepts the uploaded file and produces two outputs:

• Structured content — the document's text organized by sections, pages, and structure
• Queryable representation — an indexed form optimized for question-answering

The output format is configurable (text, JSON, or both). The parsing respects document structure — headings, lists, tables, and page boundaries are preserved, not flattened into a single text blob.

Re-parsing is available at any time. If a document was updated or the initial parse produced unexpected results, users can trigger a fresh parse that replaces the previous content.

Integration with the Document System

Document Intelligence is built on top of Roovie's existing document management system. Documents that are uploaded for any purpose — specifications attached to assemblies, reports linked to projects, reference files associated with buildings — can all be parsed and queried.

The parsing status appears directly in the document detail view:

• Not yet parsed — a "Parse Document" button appears with a description of what parsing enables
• Processing — a progress indicator shows the current stage
• Parsed — "Chat" and "Open in Workspace" buttons appear, along with page count and message count
• Failed — an error message and "Re-parse" button are shown

This means document intelligence is not a separate workflow. It is an enhancement layer on documents that already exist in the project.

What This Changes

Building projects accumulate large volumes of documentation: specifications, energy audits, commissioning reports, code compliance submittals, manufacturer cut sheets, and design narratives. These documents contain critical information, but finding specific answers within them requires reading and searching manually.

Document Intelligence changes that:

• A project manager can ask "what is the required ventilation rate for the conference rooms?" and get an answer with a citation to page 47 of the mechanical specification.
• An energy modeler can ask "what were the key findings from the last energy audit?" and get a summary with references to the audit report.
• A design agent working on HVAC sizing can access the parsed specification to verify equipment requirements before generating a system.
• A validation check can reference the commissioning plan to confirm that testing protocols match the building model.

The documents stop being static files that sit in a folder. They become active knowledge that can answer questions on demand.

What Makes This Different

Most document management in building tools is storage — files go in, files come out, and everything in between is manual reading.

Roovie's Document Intelligence is different:

• Documents are parsed into structured, queryable content — not just stored
• Questions are answered with page-level citations, not keyword search results
• Suggested questions are generated per document based on its content
• The workspace provides side-by-side document viewing and chat
• Parsed content is available as structured JSON for programmatic consumption
• AI agents can reference parsed documents to inform their design work
• Parsing is integrated into the existing document system, not a separate tool

The key is that the same documents teams already upload for project management become sources of answers — for humans through the chat interface and for AI agents through structured content.

Bottom Line

Document Intelligence turns uploaded files into queryable knowledge. Upload a PDF, specification, or report. Parse it with AI. Ask questions and get answers grounded in the document's own text with page-level citations.

For users, it means finding information in project documents takes seconds instead of manual searching. For AI agents, it means design decisions can be grounded in project-specific documentation — specifications, reports, and standards — not just building model data.

The documents teams already manage become part of the intelligence layer that drives better building design.