Managed data agents
answering where
your team works
Agents do the routine work. An engineer answers for them.
- Agents in your chat
- Your data models
- Your data layer
Three agents run on your data. One answers questions in Slack, Microsoft Teams or your own bot. One rebuilds and tests your data every day. One calls me when a person is needed.
- Operator
- Karolis Ščerbiakas, Aitoras UAB
- Answers in
- Slack, Microsoft Teams or your own bot
- Runs on
- Your data, in your own cloud
- Service
- Monthly, managed end to end
Three agents on your data. One engineer behind them.
A chatbot that queries your warehouse is easy to buy. It is wrong when your data is wrong, and nobody notices. Here each agent has one job, and every failure reaches a person who can fix it.
The Analyst answers
Your team asks in chat and gets the figure, the period, the source and how fresh the data is.
- Works in Slack, Microsoft Teams or your own bot
- Reads your data models, never raw tables
- Follows your business rules, written in plain language
- Your team marks each answer correct or wrong
The Watcher escalates
When a build fails or the Analyst cannot answer, the Watcher sends the case to me with what was tried and what is missing.
- Failed builds and failed tests
- Questions the Analyst could not answer
- Daily question limits reached
- Every case reaches me by email, not your team
The Builder keeps data current
Every day, at a time you choose, the Builder rebuilds your data models from the latest data and tests them.
- Data models in dbt, in a repository you own
- Tests that run with every build
- Column descriptions the Analyst reads
- Your dashboards stay as they are
Ask in chat. Get the number and where it came from.
An example thread. The figures are made up. Every answer your team gets has the same parts.
- Sarah
@Aitoras what was revenue last month?
- Aitoras
Revenue in September 2026 was €184,320, up 6.2% on August.
- Sarah
And how much of that came from Meta ads?
- Aitoras
None of your data models covers ad attribution yet, so I cannot answer that reliably. I have asked Karolis, and the answer will come in this thread.
Escalated to Karolis
Your business rules
Facts about your data, such as what counts as revenue, written in plain language. The Analyst reads them with every question.
Source and freshness
Which data model the figure came from and when the Builder last rebuilt it, so nobody has to guess whether it is current.
A person when it is unsure
The Analyst says what is missing instead of guessing. The Watcher sends it to me. I answer in the thread, or add the data model so the Analyst can answer next time.
The Builder rebuilds your data every day.
Your data models are built from the latest data and tested before your team starts work. If a build fails, the Watcher tells me before anyone asks.
- Fetch models
- Build
- Test
- Refresh the Analyst
- Report
What clients say about working with me.
“Karolis is the kind of freelance infrastructure engineer you want on your team - versatile, dependable, not afraid of new challenges. He upgraded our Kubernetes cluster to the latest version with auto-scaling and zero downtime, centralised our infra config, automated our deployment pipelines, and built tooling for on-demand isolated test environments. He consistently brings ideas that save costs, improve uptime, and reduce friction. Highly recommended.”
“Karolis provided the expert knowledge, personalized training and guidance I needed to bridge the gap between AI theory and real-world automation. Sessions offer condensed, high-impact knowledge that significantly improved my confidence in the subject. A must for anyone looking to fast-track their AI fundamentals and implementation strategies for any use case.”
References and further client examples are available on request.
From the first call to answers in chat.
Discovery
We go through your data, your dashboards and the questions people ask you every week. You get a proposal with the scope and the monthly price.
30-min call
Setup
You give access to your data, a git repository and your chat tool. I model your data, write down your business rules and check the answers against figures you already trust.
Before anyone asks the agents
Service
The agents answer, rebuild and watch. I take every case they send me and add data models as new questions come up.
Every month
The engineer behind the agents.
Karolis ŠčerbiakasSenior Backend Engineer & DevOps Specialist, Klaipėda, Lithuania
I run a managed data service for companies that have their data in a warehouse but no data team of their own.
The agents take the routine questions and the daily work. What they cannot do comes to me. I build and maintain the data models they read, so when a number looks wrong, a person knows why.
- Team size
- One senior engineer. No account managers, no handovers.
- Your data
- Stays in your own cloud project.
- Language model
- Claude, billed to your own cloud account.
- When an agent is stuck
- The case comes to me, and I answer in your thread.
- Claude
- Vertex AI
- BigQuery
- dbt
- SQL
- Python
- Looker
- Slack
- Microsoft Teams
- Your own bot
- Google Cloud
- Cloud Run
- Terraform
Questions before a call.
- What does it cost?
- A monthly fee, set in a proposal after the first call, once the scope is clear. The language model runs in your own cloud account, so your cloud provider bills you for it directly, with nothing added by me.
- Where does the Analyst answer?
- In Slack, Microsoft Teams, or a bot inside your own product or intranet. It answers where your team already asks questions, in a channel or in a direct message.
- Which data does it work with?
- It runs on BigQuery in Google Cloud today, with the data models in dbt. If your data is in another warehouse, say so on the call and I will tell you straight whether it fits.
- Where does our data go?
- It stays in your cloud project. The Analyst uses Claude through Vertex AI in that project, and it can only read the dataset that holds your data models. No API key is stored anywhere.
- Do we need dbt or a data team?
- No. I write and maintain the dbt models in a git repository you own. If you do not have one, I create it and transfer it to you when you ask.
- What happens when an agent cannot do something?
- The Analyst says so in the thread and says what is missing. The Watcher sends the case to me. I answer there, or add the data model so the Analyst can answer the next time someone asks.
- Are our questions stored?
- Questions and answers are deleted 14 days after the last message in a thread. Only the number of questions per day is kept, without any text.
- Is this an agency?
- No. Aitoras is one engineer. You work directly with the person who models your data and answers for the agents, with no account managers in between.
- Do you still take consulting work?
- Yes. AI engineering, backend and Google Cloud infrastructure work is still available as a project or a monthly retainer. Mention it on the call.
- Where are you based?
- In Klaipėda, Lithuania (Eastern European Time), working remotely with clients.
Tell me what your team keeps asking about the numbers.
Book a 30-minute call to walk through your data and the questions people ask. Rather write? Use the form or email me.
- We talk for 30 minutes about your data and the questions your team asks.
- You get a proposal with the scope and the monthly price.
- I model your data, connect your chat tool, and the agents start work.
- Service
- Monthly, managed end to end
Email[email protected]