Electrical contracting
From a photo of a hand-drawn plan to a priced offer in about 10 minutes
A photo of a sketch, a PDF or a spoken description goes in. A price offer, an invoice, a shop order and a go / no-go margin verdict come out. What used to take the owner two to three days per site.
An electrical contracting company. Installation work on residential and commercial sites. The owner personally priced every job.
to price a site and produce the paperwork — was 2–3 days
sites processed through the system
of the owner's week was paperwork
documents generated from one input
Problem
The owner was the bottleneck, and the bottleneck was paperwork
He was simultaneously foreman, driver, estimator, sales manager, bookkeeper and owner. A time study of his working week showed where it actually went: more than 35% on paperwork — analysing the project, pricing it, writing the offer, the contract, the invoice and the shop order. Customers waited for a quote while crews stood idle, and new work was turned down for lack of time.
Solution 1
The estimating agent
The site goes in through a web interface or a Telegram bot, in whatever form it exists: spoken aloud, typed, or uploaded as a PDF or a photo of a hand-drawn plan. The agent reads it, asks clarifying questions where the input is thin, prices the work against the company's own price list and the applicable standards, and produces the whole package.
Solution 2
A dashboard that answers what to do next
The owner's real question every morning was not «what is the status» but «which of these do I do first». One screen: today's tasks with overdue items that do not disappear until closed, a project board across seven stages from enquiry to handover, weekly goals with progress, and outstanding receivables. Priorities are derived from the company's targets, not from whatever shouted loudest.
Solution 3
Job economics you can negotiate with
A live calculator on the job card: move the price, the crew's share and the overhead, and margin and payroll recalculate as you drag. It exists for one moment — the phone call where the customer asks for a discount. The owner can now see, in that moment, exactly how far he can go before the job stops being worth doing.
2–3 days to produce documentation for one site
«Which job do I start with?» — every morning
Customers wait for a quote while crews stand idle
No idea what a job will actually earn
New work turned down for lack of time
About 10 minutes
The system shows the day's priorities
Quotes go out same-day; crews stay loaded
Margin is known before the price is agreed
The company has room to grow
What changed
instead of 2–3 days
Complete document package for a site, from a sketch or a spoken description
of the owner's time returned
He now runs the company instead of typing its paperwork
sites processed
Real jobs priced through the system, not a pilot on synthetic data
on every quote
Margin verdict before the price is agreed, not after the job is done
Under the hood
What this is built from
- Python
- Vision over hand-drawn plans and PDFs
- Voice input in the browser
- Deterministic pricing and margin engine — no LLM anywhere near the arithmetic
- Telegram bot + web interface
- Generalised version: multi-agent over A2A, pricing behind an MCP server, ADK orchestrator, guardrails on every model call
What’s next. The field solution was generalised into an anonymised, reusable system and released as open code — it became my Google × Kaggle capstone, and won the Agents for Business track.
Let’s talkabout your bottleneck
Hiring, a project, or a second opinion on an agent design — Telegram is the fastest way to reach me. I read everything myself.
mail [at] artemborisov.ru
Based in Russia (UTC+12) · open to relocation