The fastest way to stop employees from rummaging through the company wiki. With 10–20 core documents, immediate value.Documentation Index
Fetch the complete documentation index at: https://docs.teeem-ai.com/llms.txt
Use this file to discover all available pages before exploring further.
What you get
An employee asks in Slack or web chat:30-minute setup
1. Identify core documents (5 min)
Quality of the KB is everything. List the top-10 questions employees ask, then collect the documents that answer them.Examples:
- Travel policy →
travel-policy-v3.pdf - Vacation policy →
vacation-policy-2026.pdf - Expense handling →
expense-handbook.pdf - Approval flow →
approval-flowchart.pdf - Internal system manuals (Wehago, Douzone)
2. Bulk upload to KB (10 min)
Admin console → Knowledge Base → Upload → drag a folder.Per file:
Three-tier dedup prevents double-indexing if a file is uploaded twice.
| Field | Value |
|---|---|
| Department | HR / Finance / Legal etc. |
| Sensitivity | public for general, internal for HR policy, secret for executive comp |
| Tags | policy, manual, 2026 |
| Language | ko |
3. Verify with real questions (5 min)
Ask 3–5 questions employees actually ask:For each answer, verify:
- Accuracy (compared to the source PDF)
- Citation points to the right page
- Sensitivity classification is right
- For unknown questions, the agent says “no source found” honestly
4. Tighten answer policy (5 min)
For high-stakes channels (e.g. Details on the Citations page.
#legal), force citations.Admin console → Operations → Answer policy:5. Tell your employees (5 min)
The key: value only emerges if employees use it.
- Internal announcement: “Now ask @Teeem in Slack about policies”
- Five usage examples on the company wiki / intranet
- Demo at the next manager all-hands
- Number of policy questions HR receives → before/after
- Wiki page views → before/after
Document lifecycle
Policies change. The KB doesn’t need to be hand-maintained.| Situation | What happens |
|---|---|
| Policy revision (v3 → v4) | Upload new PDF → auto-supersede → v3 dropped from search |
| Policy retirement | Mark archived in admin console → excluded from search, audit trail preserved |
| Sensitivity change | public → private → change history recorded automatically |
| Department transfer | Edit department metadata → ACL re-applies automatically |
Common gotchas
Answers are too generic (low specificity)
Answers are too generic (low specificity)
The PDF’s text is too short, or information is locked in tables/images. Verify text extraction. Scanned PDFs without a text layer get OCR but with reduced accuracy.
Old versions still appear in answers
Old versions still appear in answers
Upload the new version to trigger supersede. The old version drops out of search automatically. Verify in the version history panel.
Sensitive policy reached the wrong audience
Sensitive policy reached the wrong audience
Sensitivity probably set to
public by mistake. Change to private immediately. The audit log captures everyone who accessed it.Same question gets different answers
Same question gets different answers
Possibly conflicting policies in different documents. The KB usage statistics page surfaces frequently conflicting documents — clean those up.
Next
Weekly report automation
Build scheduled reports.
Data analysis automation
Pile of Excel → insight.