Enterprise data retrieval

Enterprise Data Retrieval

Data Structure

Enterprise data lives under /root/DATA/ with two subdirectories:

• metadata/ — employee.json, customers_data.json, salesforce_team.json
• products/ — One JSON file per product (e.g., ContentForce.json)

Product JSON Schema

Each product file contains:

• slack — List of Slack messages with Channel, Message (User, text, timestamp), ThreadReplies, Reactions
• documents — List of docs with content, date, author (employee ID), type, feedback, document_link
• meeting_transcripts — List with transcript text, date, participants, document_type
• meeting_chats — List of meeting chat messages
• urls — List of links with description
• prs — List of pull requests with title, summary, reviews, user

Slack Message Schema

{
  "Channel": {"name": "channel-name", "channelID": "ch-xxx"},
  "Message": {
    "User": {"userId": "eid_xxx", "timestamp": "...", "text": "...", "utterranceID": "..."},
    "Reactions": []
  },
  "ThreadReplies": [
    {"User": {"userId": "eid_xxx", "text": "..."}}
  ]
}

Retrieval Strategy

1. Parse the question to identify: product name, entity type (document, slack, PR), and target info (employee IDs, URLs, names).
2. Load the product JSON and search relevant sections.
3. For document queries: Check author field and feedback field. Search slack for discussion threads about the document.
4. For competitor queries: Search slack messages for competitor product names. Track all participants in those discussion threads.
5. For URL queries: Search slack messages and the urls section for links matching the criteria.

Key Patterns

• Document authors are stored as employee IDs (e.g., eid_xxx) in the author field.
• Document reviewers are found in Slack discussions — look for messages between the initial document share and the updated version share.
• Competitor discussions happen across multiple channels. Search all channels for competitor product names.
• Slack conversations are sequential — messages following a topic-starter in the same channel are part of that discussion until the topic changes.

Output Format

Return answers as lists of strings (employee IDs, URLs, etc.) even for single items.