We build software, and write down how.

1. Jul 2025 - Dec 2025Patient records export and search on AzureHealthcare provider, EHR/EMR records · Direct to client
2. Jul 2025 - Dec 2025Referral intake automation with LLM summarisation in SalesforceReferral intake operations on Salesforce · Direct to client
3. 2024 - 2025Attrition prediction pipeline with calibrated models and Pega decisioningResearch prototype for service organisations · As a master's dissertation
4. Aug 2023 - Jul 2024Intake tool for a US health insurer, proof of concept to productionUS health insurer · Through an IT services firm

All seven projects →

1. 1

   Claims are backed or removed.

   Every entry on this site traces to a record of the work. There are no testimonials, no logos and no figures beyond what the records support. If a number could not be stood behind, it was left out.

2. 2

   Dates and context are stated.

   Each project says when it was done, how it was delivered, and for what kind of organisation. Work done before the studio existed is labelled as such, not folded into it.

3. 3

   Data stays inside its owner's boundary.

   Where the work touched patient records, the design kept them inside their owner's boundary: controlled storage access and sanitised exports. The attrition models were evaluated on synthetic data. The same rule applies to anything Kernwise builds.

• 15 Oct 2026A model score is not a probability until you calibrate it.7 min
• 15 Oct 2026A records export should be safe to run twice.8 min
• 26 Sept 2026A small list of things to do before you migrate to Postgres 17.2 min
• 26 Sept 2026Pool exhaustion is not database latency.3 min