Skip to main content

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.

Replace the analyst’s manual cycle of clean → aggregate → chart → conclude with one prompt. The most common pattern in finance, sales, and operations.

​
What gets automated

TaskManualTeeem AI
Merge multiple Excels + dedupe1–2 hours1–2 minutes
Per-account revenue trend chart30 minutes30 seconds
Fill quarterly report template2 hours1 minute
Extract specific fields from 100 PDFsHalf a day5 min (async)
Meeting recording → action item table1 hour2 min

​
First analysis — 5 minutes

1

1. Attach data

Drag-and-drop the file into the chat input. Or attach to a Slack channel.Supported formats:
  • Excel (xlsx, xls)
  • CSV (Korean encoding auto-detected)
  • PDF (with OCR)
  • HWP / HWPX
  • Images (OCR + visual analysis)
  • ZIP (safe extraction)
Files over 8 MB shift to async automatically.
2

2. Instruct in natural language

Specificity wins.❌ Bad: “Analyse this”✅ Good:
@Teeem In the attached account list (3 Excel files):
- Dedupe by business number
- Split active accounts (orders in last 6 months) from dormant
- Sort active accounts by revenue total descending
- Output Excel (Active sheet + Dormant sheet)
3

3. Verify the result

Teeem AI does this in one go:
  1. Safe-extract the ZIP → parse each Excel
  2. Normalise business numbers (strip hyphens, trim)
  3. Identify duplicates + merge (3 files → 1)
  4. Join with order data (if available) → revenue totals
  5. Split active/dormant by 6-month threshold
  6. Generate Excel (per-sheet, auto-width, header style)
  7. Attach to chat + summary message
PII (RRNs, phone numbers) is auto-masked before reaching the LLM, restored in the output Excel.

​
Common analysis patterns

@Teeem Merge the 5 attached account lists and dedupe.
Dedup key: business number primary, email secondary.
Output Excel.
Key: without a stated dedup key, the agent estimates one. Specify for accuracy.
@Teeem From the attached revenue data, build a 12-week trend line chart.
- 5 lines per account
- Large fonts, mobile-friendly
- PNG output
Mermaid / SVG / PNG outputs. Chart types: bar, line, pie, doughnut, scatter, heatmap.
@Teeem In the 12 attached quotation PDFs, extract:
- Quote date
- Account name
- Quote amount
- Delivery date
- Account-owner email
Excel table.
Structured PDFs hit 95%+ accuracy via OCR + pattern matching. Free-form PDFs use LLM extraction.
@Teeem In the attached 45-minute meeting recording:
- Transcribe (Korean STT)
- 5 key decisions
- Action items with owner / deadline / priority
- Recommended next-meeting agenda
Same workflow for video. YouTube link → captions are auto-pulled.
@Teeem In the attached daily orders data:
- Compute normal pattern (mean, stdev)
- Flag dates outside ±2σ
- Day-of-week pattern in outliers
- Excel + chart with outliers highlighted
Statistical analysis runs through analyze_data. For deeper work, forecast (time-series) or alert_monitor (threshold).

​
Improving result quality

Single message vs many — putting all requirements in one message yields better accuracy. Adding “and also…” mid-flow scatters context.
State explicitly:
What to specifyExample
Input format”First row is header”, “split per sheet”
Processing rules”Latest date wins on duplicates”, “skip empty rows”
Output format”Excel with Active/Dormant sheets”, “PNG 1200×800”
Length”Up to 100 rows”, “3 charts only”
Tone / language”Conclusion in Korean, max 8 sentences”

​
Common gotchas

The source likely has empty cells, total rows, or notes mixed in. Add explicit instructions: “exclude note rows”, “ignore total rows”.
Tables are images inside the PDF. OCR runs but table-recognition accuracy is 70–85%. If possible, get the raw Excel or CSV.
Per-file cap is per-tenant (default 16 MB / request). For larger:
  • Split per sheet and upload separately
  • Or zip and upload (auto-extracted)
  • Or App Pack for SQL-style queries
Without a clear dedup key, the LLM estimates. For business numbers needing normalisation (“123-45-67890” vs “1234567890”), state the rule. Or codify rules in App Pack.
Intended. PII is masked before the LLM and restored before output. If you want PII redacted in the output too, say “leave email/phone masked”.

​
Recurring analysis → make it permanent with App Pack

If you do the same analysis weekly, move it to App Pack.
  • No need to attach data each time (the app holds it)
  • Natural-language queries return immediately
  • Slack slash-command shortcut
  • Automated alerts (e.g. “stock < 100 → alert”)
User: "@Teeem this month's revenue by account"
  ↓
Agent: app_aggregate({model: "sales", group_by: "account", period: "this_month"})
  ↓
Instant answer — no attachment needed
For onboarding, talk to sales.

​
Next

Make it permanent with App Pack

Skip the “attach data” loop.

Scheduled intelligence

Run analyses on a schedule.