Software, Data & AI Engineer · Brno · remote / hybrid
AI agents
Agents that do company work
Role: Design and build of AI agents for company processes, from the first prototype to running operation.
Result: Agents for customer care, invoices, data and development, built in n8n, Make.com and code; each hands a case to a person when it is unsure.
An agent reads what comes in, works out what it is, does the routine part and hands the rest to a person with the work already prepared. Each agent has set limits on what it may touch, how it decides and who checks it.
The agents
- Customer care: an agent that reads incoming mail, sorts and summarises it, routes it to the right team and drafts a reply for the operator.
- Invoices: AI that reads supplier invoices and writes the data into the CRM.
- Data: an agent that answers questions over the data warehouse.
- Coding: agents that write and check code, under tests and independent review.
How they are built
- Typed decisions: where a wrong choice costs money, TypeSafe Jev returns a calibrated probability instead of free text; a clear margin lets the agent act, anything less goes to a person.
- Approved knowledge first: agents answer from approved, versioned sources, and permissions are applied before anything reaches the model; a fact that cannot be verified is not claimed.
- Limited tools: agents get only the tools they need, read-only where possible: mcp-sql-agent for SQL Server and PostgreSQL, powerbi-mcp with 45 Power BI tools.
- Autonomy is earned: a process runs on its own only after 30 correct cases in a row and the process owner's approval, with a kill switch.
- Checked and logged: tests, cost budgets and an audit trail of every decision; e-mails and attachments are treated as untrusted input.
Stack
n8n · Make.com · TypeSafe Jev · LLM gateway · MCP · PostgreSQL + pgvector · .NET · Python · Claude Code