A conversational inventory management system that understands natural language, handles manager approvals, and maintains complete audit trails â all through Telegram
Engineer
Sends message
Telegram Bot
+ n8n + OpenAI
Google Sheets
Inventory + Logs
Designer & Builder
4-6 weeks
Telegram (Mobile/Desktop)
n8n, Telegram API, OpenAI, Google Sheets
Live in Production
Imagine you're an engineer on a construction site and you need to check if there are any GRE flanges in the warehouse. Instead of walking to the store and checking a spreadsheet, you just text a Telegram bot:
đĻ INVENTORY CHECK
Item: GRE FLANGE 2 Inch
Current Stock: 45 Pcs
Location: Shelf B3
Use: Take 10 | Add 5
The bot understands natural language (you don't need exact commands), handles the entire workflow from checking stock to manager approvals, and maintains a complete audit trail. It's essentially a smart inventory assistant that lives where the team already communicates.
No app to download, no login to remember. Users interact through Telegram â a tool they already use daily.
Type "gre flan" or "check upvc pipe" â the system understands. No need to memorize exact commands.
Managers approve requests with one tap on their phone â no laptop needed, even when on-site.
Check inventory, request items via TAKE command, view transactions
Comfortable with phones; prefer mobile over spreadsheets on-site
Approve/reject requests, add new stock via ADD command
Mobile-first; often moving around, don't use Excel much
Oversight, transaction history review, direct Sheets access
High tech comfort; comfortable with spreadsheets and data analysis
A construction/engineering company's inventory tracking was a mess of manual processes that wasted time and created accountability gaps.
đļ Physical Warehouse Trips
Engineers had to walk to the warehouse just to check if items were available
đ Outdated Spreadsheets
Stock levels tracked in Excel files that were often outdated by the time someone looked at them
â No Audit Trail
No clear record of who took what, when, and for which project
đģ Laptop-Dependent Managers
Managers had no visibility into real-time inventory without opening Excel on a computer
đī¸ Project Conflicts
Material allocation for different projects tracked verbally or not at all
â ī¸ Emergency Shortages
Low stock situations discovered only when someone physically ran out of an item
Hours/week wasted on manual tracking
Unknown materials taken without records
Project delays from unexpected shortages
Operations Manager
"I need visibility and audit trails for material usage"
Engineers
"I want to check stock without walking to the warehouse"
Store Managers
"I need to approve requests from my phone when I'm on-site"
n8n
Workflow Automation
Telegram Bot
User Interface
OpenAI
Natural Language
Google Sheets
Database
"gre flange 2 inch"
Search and display inventory item details
"take 10"
Request items (triggers manager approval)
"add 50"
Restock items (triggers manager approval)
"list low stock"
Show all items below minimum stock level
"sheet"
Get direct link to Google Sheets
"2" (after search)
Choose from multiple matching items
Found 9 matching items:
1. DUAL CONCENTRIC REDUCER 1x0.5" - Stock: 9
2. DUAL CONCENTRIC REDUCER 2x0.5" - Stock: 8
3. DUAL CONCENTRIC REDUCER 3x2" - Stock: 14
Reply with a number (1-9) to see details
đĻ INVENTORY CHECK
Item: DUAL CONCENTRIC REDUCER 3x2"
Stock: 14 Pcs
Location: Shelf C2
Use: Take 5 | Add 10
âŗ REQUEST SUBMITTED
Item: DUAL CONCENTRIC REDUCER 3x2"
Quantity: 5 Pcs
Transaction ID: T134
AWAITING MANAGER APPROVAL
â APPROVED â T134
DUAL CONCENTRIC REDUCER 3x2"
Qty: 5 | New Stock: 9
Collect from store.
Used OpenAI to parse messages so users can type "gre flange 1 inch" or "check upvc pipe" instead of memorizing exact syntax. This dramatically improved adoption since engineers don't need to remember command formats.
Even for ADD commands (restocking), kept the approval workflow as a "double-check" system. There was team debate â one engineer said it was unnecessary, but the Operations Manager valued the accountability. The 2-second approval click prevents costly inventory errors.
Chose Google Sheets over a traditional database for transparency â managers can still open the spreadsheet directly when needed, and it's familiar to everyone. Also allows for easy manual corrections if needed.
After viewing an item, users can just type "take 5" without repeating the item name. The system remembers what they were looking at, creating a natural conversational flow.
Tries exact matching first, then falls back to partial matching. "gre flan" still finds "GRE FLANGE". Handles typos and abbreviations gracefully.
When a manager approves a request, the system re-checks current inventory before processing. If stock changed since the request was made, it rejects with a clear message instead of going negative.
For items measured in meters (like pipes), users can request "take 0.3" for 0.3 meters. Required special handling in the AI prompt to preserve exact decimals.
Engineers can cancel pending requests before approval. The system updates the manager's message to show it was cancelled and logs the cancellation with timestamp.
Telegram Bot â Check / Select / Take Flow
Guided conversation flow for submitting attendance actions
Manager Approval Interface
Inline approve / reject actions directly within Telegram
| Sheet Name | Purpose | Key Columns |
|---|---|---|
| INVENTORY | Master item list | Item_ID, Material, Description, Size, Quantity, Unit, Location, Min_Stock |
| TRANSACTION_HISTORY | Audit trail | Trans_ID, Timestamp, Item_ID, Quantity_Change, Old/New_Qty, Type, Employee, Status, Approved_By |
| RESERVATIONS | Project allocations | Reservation_ID, Item_ID, For_Engineer, Project_Code, Tagged_Qty, Status |
| USERS | Team members | Telegram_Username, Display_Name, Role (engineer/manager), Chat_ID |
Transactions Logged
Hours/Week Saved
Items Tracked
Monthly Cost
OAuth token expiration caused recurring manual work until I switched to Google Service Account credentials. Lesson: use non-expiring auth methods for automated systems.
Session caching is useful, but always re-validate from the database before executing transactions. Stale data in approval workflows can cause inventory issues.
Natural language processing greatly improves adoption, but requires careful prompt engineering. Edge cases like "take 0.3" were being rounded until I explicitly told the AI to preserve decimals.
Low Stock Alerts
Proactive notifications when items fall below minimum levels
RETURN Command
Allow engineers to return unused items back to inventory
Full Reservation System
Block reserved stock from general TAKE commands
Enhanced Search
Better handling of description-only searches