Kraken.

01 / Context

The platform

Electricity settlement turns meter data into the consumption records used by the energy industry. At Link’s scale, an unbounded query or repeated lookup can make an operational page unusable or push background work past its queue timeout.

02 / Responsibility

Selected work

I work on Link, Kraken’s electricity settlement platform, using Python, Django and PostgreSQL. My work ranges from fixing slow queries and reclaiming database storage to building job execution and sign-in workflows.

• Making large datasets usableRemoved timeouts from a view spanning approximately 26 million records. A separate message-query benchmark fell from about 63 seconds to 16.5 ms.Read the case study →
• Operational commands, from request to resultBuilt a permissioned Kubernetes Jobs workflow with tracked state, captured output, duplicate-execution guards and reconciliation.Read the case study →
• Fixing a queue consumer’s database bottleneckReplaced per-meter lookups with batches of 1,000. For 100,000 meter points, the query model falls from 200,001 queries to 201.Read the case study →
• One sign-in, two organisationsImplemented and tested OIDC sign-in with automatic tenant routing, then added default UI protection and signed access to a related service.Read the case study →

03 / Approach

How I approach the work

I focus on the work each request actually needs: bound database reads, batch repeated lookups and remove redundant indexes. For operational tools, I make permissions, lifecycle state and failure handling part of the workflow. The four featured case studies show those decisions in detail.

1. 01

   Understand the failure

   Trace a slow page, repeated queue message or incomplete operational workflow to its cause.

2. 02

   Change the work

   Bound reads, batch queries and model execution state explicitly.

3. 03

   Verify the result

   Use query plans, regression tests and end-to-end verification commands appropriate to the change.

04 / Result

Results across the platform

Eliminated timeouts on a view spanning approximately 26 million records and reclaimed approximately 1.43 TB by removing 11 redundant indexes across two production PostgreSQL estates. I also delivered tracked command execution through Kubernetes Jobs, bulk settlement queries and SSO for two organisations. Related validation work blocked approximately 483,000 invalid updates per month across two enterprise clients.