Industry: Construction (Iowa, US)
A contractor that stopped doing data entry
Supplier invoices, concrete deliveries, receipts and jobs from five different systems, pulled into one database with daily Telegram alerts.
The code is private and the client is described without their name.
The problem
The company's information lived in five platforms that didn't talk to each other: the job system, the concrete delivery system, crew time tracking, a builder's portal and the inbox where invoices arrived. Knowing what got done, what still needed billing and what was billed wrong meant opening everything and comparing by hand.
The system
- Invoice PDFs that arrive by email are read by AI: vendor, date, job and amounts are saved with zero typing.
- Receipt photos go to a Telegram bot; AI pulls the details and the person just taps to confirm.
- Every few minutes the system pulls concrete deliveries, jobs and schedules from the other platforms into one database.
- It cross-checks everything and flags what nobody would catch by eye: work done but not invoiced, invoices with the wrong address, deliveries with no job attached.
- Every morning it sends the day's plan and open items on Telegram.
The result
- 5 systems connected in a single database
- 7,859 line items from 1,793 jobs synced
- 3,095 job-site photos stored in a private archive
- Daily plan on Telegram at 8:00 am, with 3 items max
This is the largest system I’ve built and it keeps growing: it started by reading receipts and now connects most of the operation. The rule from day one was that nobody had to change how they work — the system adapts to the tools the crew already used.
Built with
- Node.js
- Supabase (PostgreSQL)
- OpenAI
- Telegram
- Render