Roster

Your machines. Their costs. One AI agent.

A construction company with equipment in Bulgaria and abroad. Fuel, rent, tolls, vignettes, repairs — tracked in one place, asked in plain language, confirmed by your people before anything is saved.

Scope Architecture and demo — not a finished product
First version Telegram + portal

Flow

How it works

Standard invoices, receipts, and handwritten documents. Fields: amount, sender, receiver, VAT. A person confirms before data is stored.

1

Send

Photo or file in Telegram.

2

Read

OCR + agent extracts the fields.

3

Confirm

Overview on screen. Fix if needed.

4

Store

Saved by machine, place, driver, period.

5

Use

Dashboard, questions, then send approved invoices out.

Architecture

How the pieces connect

CHANNELS Telegram Upload, confirm, ask Portal Dashboard, filters, users AI AGENT New agent · OCR · extract fields · wait for confirmation · answer cost questions PLATFORM · new module Backend APIs, validation, send-out Database Machines, costs, invoices Frontend Confirm UI, dashboard OCR Photos and files AFTER APPROVAL Client platform, email, or another channel

Stack

Technologies

Telegram

Channel for the agent. First version uses this instead of a separate mobile app.

React

Portal and dashboards. Confirm screen, filters by machine, country, city, driver, period.

NestJS

Backend on our existing platform. New APIs for this module.

PostgreSQL

New tables for machines, sites, drivers, documents, extracted fields, costs.

AI + OCR

Read documents and run the agent. After confirm, questions hit stored data — not the same photo again.

Docker

Same hosting setup we already run. This module is added there.

New work

What we need to build

Build

Backend

APIs, confirmation, validation, send approved invoices out.

Build

Frontend

Confirm / correct screen. Dashboard with fleet filters.

Build

Database

Machines, locations, drivers, invoices, cost types.

Build

OCR

Standard invoices, receipts, handwriting (lower accuracy).

Build

AI agent

New agent: upload, extract, confirm, answer cost questions.

We reuse the existing platform (logins, portal, dashboards, hosting). The Roster module is new.

~2 months · 2 people, then stabilize

Milestones

Month 1

Foundation

  • Database tables
  • Backend APIs
  • OCR on sample invoices
  • Agent receives a document and returns fields
Month 2

Operate

  • Confirm / correct
  • Dashboard + filters
  • Costs by machine, place, driver, period
  • Send approved invoices out
After

Stabilize

  • Test on real documents
  • Corrections improve the next reads
  • Later: dedicated web or mobile if needed

Shortcuts

Presenting

→ Space next · ← previous
Home / End · F fullscreen · ? this panel