Serhii Khalabuzar
All projects

PVC and aluminium window installation and glazing

Work tracking and an AI assistant for a window installer

Crews send their reports in Telegram, the bot checks the data and produces a ready payment statement in under a minute, and an AI assistant answers questions about site progress.

Client
Window installation company
Format
Commercial case
Stack
n8n, Google Apps Script, Google Sheets
  • < 1 minper payment statement instead of 5–30 minutes
  • 2automatic checks on every report
  • 3AI assistant tools for the manager
n8n: report, checks, statement and AI assistant in one workflow

The problem

The company installs and glazes PVC and aluminium structures and does waterproofing and vapour barriers on residential sites. Crews report their work once a week as a text message in Telegram, and those reports feed the records for every single window unit and the payment statement for each period.

Reports used to be copied into a spreadsheet by hand and checked for duplicates, and the area of each unit was looked up manually and multiplied by the rate. One statement took 5–30 minutes and often contained mistakes. To find out how much work was left, the manager had to open the spreadsheet himself.

The solution

Automation built on n8n, a Telegram bot and Google Apps Script. The crew just writes a report in the chat and the system does the rest.

  1. Report to data. An LLM reads the free-form report (site, period, type of work, rate, risers and floors) and turns it into a list of specific units.
  2. Two checks. Every unit is checked for repeats and against the site’s master list. If something does not match, the bot stops and asks for confirmation in Telegram. No statement is created from wrong data.
  3. Automatic statement. Apps Script calculates area × rate, fills a Google Docs template and sends the finished .docx to Telegram.
  4. AI assistant. Any message that starts with «Ai» goes to a separate agent with three tools: how much is done and how much is left (in units and m²), the status of a specific unit by riser and floor, and statistics for a chosen period. The agent always takes its numbers from the site’s real data.

Each site is its own spreadsheet and script, so a new site is connected by cloning the template, with no code changes.

The result

Before After
Payment statement 5–30 min under 1 min
Data checks manual automatic, 2 checks
Stats for the manager open the spreadsheet ask in Telegram
New site manual setup clone the template

Work is tracked across 4 types of work for two materials: PVC and aluminium.

How the work went

How work on this project went, from understanding the problem to launch and improvements.

  1. 01

    Discovery

    Mapped how crews report and how those reports become a pay sheet: copying into a spreadsheet, looking up the area of each unit, multiplying by the rate. One pay sheet took 5–30 minutes and often had mistakes.

  2. 02

    Solution design

    Designed the data structure: a separate sheet and script per site, a list of units by riser and floor, and rates for 4 types of work. Decided where the system must stop and ask a person.

  3. 03

    Build

    Built an n8n workflow with a Telegram bot: an LLM parses the report text, Apps Script calculates area × rate and fills the pay sheet template in Google Docs. Added a separate AI assistant with three tools.

  4. 04

    Testing

    Tested parsing on reports in different formats, on duplicates and on units that are not on the site list. Checked that no pay sheet is created from bad data.

  5. 05

    Launch

    Crews keep writing reports in Telegram as before, so the team had no new tool to learn. The finished .docx arrives in the chat in under a minute.

  6. 06

    Optimisation

    A new site is added by cloning the template, with no code changes. The manager gets stats by asking in Telegram instead of opening the spreadsheet.