Selected work
Five cases from the people behind Tamarillo
Approach. Architect, lead, build. Kjetil co-designed the solution and led the 2 to 3 engineers who built it. Live since 2022. Tamarillo is set up the same way.
| The problem | What we did | The outcome |
|---|
| Two countries asked for the same KPI would give two different numbers. Each office had its own consultants, its own pipelines, and its own definitions. | Set one definition per KPI and built one shared model for calls, emails, events, campaigns, and consents. | Core markets can be compared side by side. |
| Dozens of markets were paying separately to build and maintain the same reports. | Moved the countries onto the shared model, with room for local analysis on top. | The consultant-built country solutions were shut down. |
| Data on healthcare professionals is sensitive, and a market may only see its own. The usual answer is a separate copy per country. | Kept one copy for everyone. A single list maps user groups to 70+ markets, and the access rules are generated from it. | A rep in Spain and an analyst in the US query the same tables, and each sees only their own rows. |
| Calculations that sit inside dashboards have to be redone every time the BI tool changes. | Kept every definition and calculation in the data model, so a dashboard only has to display the numbers. | Lundbeck switched from Qlik to Power BI without rebuilding the model. |
Result. One set of sales numbers for the core markets, and no more paying for the same report in every country.
Technology. Snowflake, dbt, Fivetran, Veeva CRM, Salesforce Marketing Cloud
Approach. Architect, lead, build. Lundbeck's commercial data scientists designed the logic, and Kjetil's team made it run in production. Live since 2022.
| The problem | What we did | The outcome |
|---|
| The data scientists had worked out which doctors a rep should contact next, but the logic only existed in their own analysis and did not reach the reps. | Rebuilt the logic as production pipelines that run automatically every day. | 9 types of recommendation are live, and new ones are still being added. |
| The information needed was spread across systems: visits in the CRM, emails in the marketing platform, sign-ups on websites, market data from vendors. | Joined it into one picture per healthcare professional, refreshed daily. | A rep is prompted to follow up after an event, collect a missing consent, or revisit a plan. |
| Reps were unlikely to open a separate tool to look for recommendations. | Synced the suggestions into the CRM as normal suggestions, each with a short reason in the rep's own language. | Reps did not have to learn a new tool. The suggestions appear in the CRM they already use. |
Result. A reported double-digit lift in engagement from campaigns driven by the recommendations.
Technology. Snowflake, dbt, Python, Hightouch, Veeva CRM
Approach. Architect, lead, build. Built together with Lundbeck's supply chain analytics team, who owned the business questions. Live since 2025.
| The problem | What we did | The outcome |
|---|
| New questions needed a specialised SAP consultant and 3 to 6 months of waiting for a report. | Copied the core SAP data onto the data platform, so reporting no longer depends on SAP's own tools. | The supply chain analytics team builds its own dashboards and answers new questions itself. |
| SAP spreads related data across many tables, and its tools build each report separately. One extra field is a new project. | Modelled the data once into plain tables for materials, plants, stock, and movements. These are documented, tested, and reused by every report. | One extra field is a small change, and every number traces back to its source in SAP. |
| Stock-outs were a recurring and costly problem, and the analysts could not get to the data needed to predict them. | Gave the analysts the full history, next to data from the other manufacturing systems, in a place where they can run their own models. | The team built its own stock predictions and reduced the stock-outs. |
Result. The analytics team serves itself, and used that to reduce a recurring and costly problem.
Technology. Snowflake, dbt, SAP, Fivetran
Approach. Architect, lead, build. Kjetil co-designed the platform as Senior Solution Architect and ran it as Head of Data & Platform Engineering, leading 7 in-house engineers and 3 to 10 external consultants. 2022 to 2025.
| The problem | What we did | The outcome |
|---|
| Every data request started from zero, with new consultants, a new budget, and a new version of the numbers. | Built one shared platform and took it into use one business problem at a time: sales first, then manufacturing and supply chain, then research. | 70+ source systems, 3,000+ data models, and 1,000+ users. New work starts from what is already there. |
| Researchers prefer local copies and their own scripts, which stops working when a dataset reaches terabytes. | Moved the large research datasets onto the platform. | Researchers analyse the full datasets on the platform instead of on local copies. |
| Without visibility into who uses what, cloud costs are hard to control. | Ran the infrastructure centrally, tuned to each type of work, and showed every business area the cost of its own usage. | Each area saw its own costs, and badly written queries were caught and corrected. |
| A pharma company must be able to show who changed what, and when. | Qualified the platform and the solutions on it, and put everything under version control, with tests and documentation. | Run in a validated state under GAMP 5, with access control and audit trail in line with 21 CFR Part 11 and ALCOA+, and within Lundbeck's ISO 27001 information security management. The platform supported internal audits, supplier audits, and regulatory inspections. |
Result. New solutions build on what already exists instead of starting over.
Technology. Snowflake, dbt, Dagster, Terraform, Fivetran, Azure, Python
Cool Sheep AI
Developed by Awesomity Lab for NSFO, an international organisation for sheep and goat breeders based in the Netherlands.
| The problem | What they did | What it shows |
|---|
| NSFO wanted to estimate the weight of a lamb from an ordinary photo. | Developed a machine learning system that takes a standard colour image and predicts the animal's weight. | The team builds applied machine learning, and has done it for a European client in a specialised field. |
Technology. AI/ML, computer vision
More work from Awesomity Lab
| Client | What they built |
|---|
| OX Delivers (UK) | A pay-as-you-go platform for electric vehicles that registers usage, optimises logistics, and collects supply chain data. |
|---|
| Move by Volkswagen (Rwanda) | The country's first large-scale mobile app, with more than 100,000 rides in under a year. |
|---|
| Harambee (South Africa) | Candidate data moved from a large number of Excel files into a platform with reporting and real-time data. |
|---|
