%402x.svg)
In one day at your office, your project managers, engineers and analysts take one data project from a stakeholder transcript to a merged pull request with Claude Code, and keep the setup.
The workshop at a glance
One day at your office, 09:00 to 17:00, five modules
Anyone in the data team, up to 12 people: project managers, data protection officers, data engineers, analysts
From EUR 9,000, fixed in the proposal. Licences for the day included
NordicMart, a synthetic online retailer, or an anonymised mirror of your data. Your production systems are untouched
Michał Dębski runs the day
Co-founder, Managing Director and Solution Architect. 15 years in data and AI; delivers in English, Polish or French.
The day runs on its own stack and dataset, so the whole team can join without access to your production systems. Data engineers only? See the data engineering workshop →
What agent-assisted delivery did at a global manufacturing group, what it can be worth for your team, and what we commit to
A shared Claude Code repo with three MCP servers (filesystem, DuckDB, Azure DevOps), skills, and a markdown knowledge base the agent reads on demand.
A configured repo with team and personal workspaces
A three-page stakeholder transcript becomes epics, user stories and a source-to-target map, with a data-protection check of every personal-data field, and the team cuts sprint one.
A populated backlog, a DPIA-lite note and the ticket-writer skill
Bronze, silver and gold models in dbt, a legacy margin script refactored, deduplication matches reviewed.
dbt models with tests, docs and lineage
A Streamlit dashboard that answers questions asked in plain English, behind three guardrails: 30-second timeout, SELECT-only access, PII masking.
The dashboard and a findings log
An agent takes an Azure DevOps ticket through plan, code and tests to a pull request, and the team reviews and merges it.
Merged pull requests and a governance and cost playbook
Stack for the day
Claude Code, MCP servers, dbt, DuckDB, Airflow, Streamlit and Azure DevOps, in Docker on each laptop. The team also keeps the prompt library, the knowledge base and a one-page runbook. Coffee at 11:30, lunch at 13:45; the hour-by-hour plan is in the brief.
For the sponsor
A cycle-time baseline from the day, a written scope for the next 90 days, an AI governance starter pack (DPIA-lite template, scope boundaries, controls, audit hooks, escalation) and an answer for the board on how the data team uses AI.
8
EUR 180,000
40% (assumption)
60% (assumption)
About EUR 345,000
The model: 8 × EUR 180,000 × 40% × 60% = EUR 345,600. In the scoping call we redo the sum with your own figures, starting with how many days one change takes your team from spec to merged model. If they show less than 4x ROI, we tell you not to book.
Scoping call, 30 minutes
We walk your stack and the work your team does, and choose the case the day is built around.
Proposal within 48 hours
Fixed fee, scope and date. We work under our MSA or sign yours.
One week before
We tune the environment to your case and send a 15-minute pre-read with the install steps.
The day, 09:00 to 17:00
Five modules at your office. The runbook stays with your team.
After the day
Your team runs the pattern itself with our Teams support, or continues into a 90-day implementation sprint at a fixed fee, or into a monthly capacity pack with a named Astral Forest engineer.
No. The stack runs in Docker on each participant's laptop, with the synthetic NordicMart dataset or an anonymised mirror of your data, so your production code and data stay outside the engagement. Models run in enterprise mode, where inputs and outputs are not used for training. A security one-pager covering data handling, access, agent scoping and subprocessors is available under NDA.
You do, from the end of the day. The repo, agent setup, skills, prompts, knowledge base and runbook are transferred to you under the engagement.
Yes, and we confirm it in the scoping call. Common swaps are Jira for Azure DevOps, GitHub Copilot next to Claude Code, Snowflake or BigQuery for DuckDB, and Power BI or Looker for the dashboard. Bring an anonymised or synthetic mirror of your data and we build the modules around your case; otherwise we use NordicMart.
On-site at your office, for up to 12 people from one company, because above 12 each person stops getting hands-on time with the agent. For a larger team we run the day twice. Inside the EU and the UK, travel is included in the fee; elsewhere, including the US East Coast on request, travel and one prep day are billed at cost. We deliver in English, Polish or French.
Yes. The SOW sets the success criterion: by the end of the day the team has merged at least two agent-drafted artefacts into the workshop repository. If we miss it, we run a second session at no extra cost. If the pattern has not stuck 30 days later, we tell you so and what we would change.
Book 30 minutes with Michał Dębski to walk your stack and agree the case the day is built around. If you want to hear from a client before you sign, we set up a reference call after the scoping call.
