Case Study · Manufacturing

Production Planning Agent

How an AI agent helped an aerosol packaging factory make production decisions in seconds — right from Telegram.

Client: Aerosol Factory →

Client

Aerosol Factory is a Russian manufacturer of aerosol tin packaging and ready-to-use aerosol products. The facility is located in Vladimir Oblast in a special economic zone, 90 km from Moscow.

Full production cycle: formula development, contract filling, testing and certification. Their own OKKE brands cover automotive care, household chemicals and disinfectants. Minimum production run: 10,000 units.

Location

Vladimir Oblast, SEZ

Min. batch

10,000 units

Cycle

Full — from formula to finished product

Brands

OKKE AUTO, OKKE PROF, OKKE SANI

Challenge

Data scattered across systems

Warehouse stock, current orders and product specs were stored in different places. Making one production decision required manually gathering information from multiple sources.

Planning took time

Before answering "how much should we produce", the planner had to check several spreadsheets and systems. This slowed down operational decisions.

No single access point

Different roles — sales manager, technologist, procurement — accessed data differently. There was no single tool that could understand a request in natural language.

Solution

We built an AI planning agent accessible directly from Telegram. The agent is connected to production data through a set of tools: product catalog, warehouse inventory, current orders.

The planner writes a request in natural language — for example, "How many units of product X should we schedule for production?". The agent queries relevant data, analyses the situation and returns a specific recommendation.

The workflow is orchestrated through n8n: receive message from Telegram, call LLM, call data tools, form a response, send it back to the chat.

01Request in Telegram
02LLM analyses the task
03Agent queries data
04Recommendation to chat

Demo

Telegram — planner interface

n8n — workflow orchestration

Result

Seconds instead of hours

A production volume recommendation is generated instantly — the agent gathers the required data from different sources on its own.

Natural language

No system training required. The planner communicates with the agent the same way as with a colleague — in Telegram, in a familiar interface.

Passed validation

The prototype passed response quality evaluation. The agent is ready to connect to real data channels and industrial use.

Stack

Telegram

Interface for the planner — familiar messenger, no additional tools needed

n8n

Orchestration of the entire workflow: receive request, call LLM, data tools, response

LLM

Reasoning engine: understanding the request, selecting tools, synthesising the recommendation

Data tools

Access to product catalog, warehouse inventory and current orders

Want a similar solution?

Get in touch — we'll discuss your process and show how an agent can fit into your operations.