KMVSG cases

Projects and pilots

Explore messaging, request workflows and 1C reconciliation. Each case states what was verified and where the result ends.

AutomationWeb servicesIntegrationsInfrastructure

Examples of our work

What we have built and verified

Messaging, request workflows and 1C reconciliation: verified results and the limits of each test.

Live pilot · MAX and Open Channels

20 MAX accounts: messages and PDFs in Bitrix24

Challenge. Bring together messages from 20 MAX accounts and test a pilot Open Channel route with replies from the normal agent interface.

The setup includes 20 MAX accounts with separate connections. On 22 October, an agreed pair was used to test the dedicated pilot routes. Real text and agent replies, PDFs in both directions with matching SHA-256 checksums, and delivery and read events were verified. A restart and event replay produced no additional MAX sends.

The route was tested with an agreed pair of accounts; existing production channels were preserved. Photos, voice, video, edits, deletions and a first outbound message to a new contact have not yet passed acceptance testing for Bitrix24. Separate 50 MiB tests with a simulated channel are outside this result. Automatic startup after reboot and a production SLA require separate agreement.

  • 20 accounts
  • Separate routes
  • PDFs both ways
  • Duplicate controls
How we verified messages and files

Pilot · Cost-estimate requests

Ten bot questions, one cost-estimate task

Challenge. Collect cost-estimation inputs, identify technical work and prepare the handoff to assignees.

On 23 October, we checked ten questions, saving and resuming a draft, the summary and confirmation. One control task was created with the correct project, assignee and two technical observers. Replaying the dispatcher checked the same task; all 121 release tests passed.

The live check used the owner’s personal authorization context and an explicitly labelled pilot task. The coordinator and technicians still need to complete team acceptance testing through the application. The bot collects requests; a numerical calculator and automatic stage restrictions are not implemented. No orders or payments were created during testing.

  • Saved draft
  • 10 questions
  • Technical work
  • Creation verified
Explore the request workflow

Completed diagnosis · Read-only

1C Accounting and UNF: 23 operations explain the bank difference

Challenge. Find the cause of different bank balances in two 1C databases and prepare a plan to restore the original operations.

On 23 October, we checked “opening balance + movements = closing balance” for seven accounts. The current difference was fully explained by 23 missing movement-generating documents: 18 withdrawals, three receipts and two transfers. The interface confirmed that native exchange had been disabled after earlier errors.

We examined active movements, not just posting flags. A posted document without bank movements was excluded from the plan; both sides of transfers were retained. Documents and settings were not changed: this stage delivered diagnosis and a plan. Restarting exchange and restoring missing operations require separate work and repeat reconciliation.

  • 7 accounts
  • 23 operations
  • Native exchange
  • No database writes
Read the bank-movement reconciliation

Other implementation scenarios

What else we can build for your team

These examples describe possible solutions and their components. Scope, timing and deliverables are agreed after discussing your project.

01 Support and integrations

Website, email and messenger requests in one queue

Instead of forwarding messages between chats, the operator gets one conversation with history, ownership and status.

Challenge

Requests arrive through separate channels, lose context and are manually copied into spreadsheets or passed between employees.

What we build

  • Deploy a self-hosted Chatwoot helpdesk and connect the required channels.
  • Configure teams, roles, routing, APIs and webhook events.
  • When needed, link conversations to a structured Baserow work register.
Working result

The team works from one interface, sees the conversation history and transfers ownership without losing context.

02 Portals and data

An operational seller workspace with metrics and deviations

Stock, prices and operational metrics are presented in an interface that highlights changes requiring action.

Challenge

Data is spread across systems and spreadsheets, so deviations are found late and reconciliation takes time.

What we build

  • Connect sources and normalize the metrics into one structure.
  • Build a workspace with filters, trends and operational statuses.
  • Add thresholds and notifications for changes that need attention.
Working result

The team replaces manual reconciliation with one working view and notices critical changes sooner.

03 Catalog and API

Preparing and uploading product cards to Kazakhstan’s national catalog

Product cards follow a controlled path from source tables to catalog submission and status tracking.

Challenge

Russian and Kazakh names, classification codes and bulk submissions must be prepared without repeating manual work.

What we build

  • Read source data from Excel or an internal system.
  • Normalize fields, classifiers and bilingual product names.
  • Submit cards through the API and expose processing errors and statuses.
Working result

Uploading becomes repeatable: problematic items are visible before the next submission and processing history is retained.

04 Corporate communications

A work messenger controlled by the company

Rooms, employee accounts, access rules and the history of work discussions remain inside the corporate system.

Challenge

Work discussions depend on personal accounts, while the company lacks unified control over access, rooms and data storage.

What we build

  • Deploy Matrix and Element on the agreed infrastructure.
  • Configure spaces, rooms, roles and access rules.
  • Add notifications, required bridges, backups and user instructions.
Working result

The company manages work accounts and spaces, while employees get one clear communication channel.

05 DevOps and infrastructure

Predictable delivery of web services from GitHub

A release stops being a collection of manual commands: build, publishing and verification follow one reproducible route.

Challenge

Manual deployments are hard to reproduce and verify, and a command error can affect neighboring services or configuration.

What we build

  • Describe the application and dependencies with Docker Compose.
  • Configure GitHub delivery, domains, TLS and isolated environments.
  • Add backups, health checks, logs and a clear rollback procedure.
Working result

Changes follow the same verifiable path, with release state and rollback options defined in advance.

06 AI and automation

An assistant for recurring questions and internal policies

AI works within defined boundaries: it uses approved materials, follows a scenario and hands unusual situations to a person.

Challenge

Employees and clients repeat the same questions, while answers depend on who is available and where the latest instruction is stored.

What we build

  • Prepare an approved set of documents, instructions and answers.
  • Connect the assistant to a work channel and define behavior rules.
  • Add memory, logging and escalation of complex requests to an employee.
Working result

Routine requests are handled consistently, while the team joins where human judgment is actually needed.