The engineering lab

We build things we’d be proud to run.

Honest label: everything here is independent work — products and experiments we built and run ourselves, not client projects.

01K8LensOpen source
02MinutesPrivate beta
03CostlensTechnical experiment

01Independent product · Open source

K8Lens

A read-only Kubernetes dashboard focused on diagnosis — it explains what broke and why.

The problem

The official Kubernetes Dashboard was archived in January 2026, and most replacements show what’s running rather than what’s wrong.

Engineering decisions

Diagnosis first. Plain-language diagnostics, a cluster health score, and blast-radius analysis with dependency graphs.
No agents, no Prometheus. Metrics come from Metrics Server, with a Kubelet Summary API fallback. One container image, local or in-cluster.
Read-only by design. Needs only get, list and watch RBAC. Works with kubeconfig, bearer token or service account on Kubernetes 1.24+.

Visit k8lens.dev GitHub

02Independent product · Private beta

Minutes

A meeting analyzer that turns recordings into decisions, owners and dates.

The problem

Decisions made in meetings vanish into recordings nobody rewatches. Follow-ups slip.

Engineering decisions

Structured extraction. LLM output is validated against a schema — every item links to a timestamp.
Queue-based pipeline. Transcription and analysis run as idempotent jobs; retries never duplicate.
Privacy by default. Audio is deleted after 24 hours. Transcripts stay in your workspace.

Try a sample meeting Architecture notes

minutes.codescapes.dev

Minutes meeting summary screenshot

03Technical experiment

Costlens

Daily cloud cost attribution, down to the team and the Terraform module.

The problem

Teams see the cloud bill a month late, as one number nobody owns.

Engineering decisions

DuckDB on cost exports. Billing exports are queried in place — no warehouse to run.
Tag-based attribution. Maps spend to Terraform modules and owning teams automatically.
Simple anomaly alerts. Rolling baselines per service flag spikes the same day.

Interactive demo Write-up

costlens.codescapes.dev

Costlens cost explorer screenshot

Why we keep a lab

We don’t recommend what we haven’t run ourselves.

Tested before recommended

New tools earn a place in client work only after they’ve held up in ours.

We feel the pager

Running our own products keeps us honest about what operating software really costs.

Judgment stays current

Engineers rotate through the lab, so experience from it reaches every engagement.

Like how we build? Let’s build yours.

The same engineers who built these products can look at yours. Start with a 30-minute call.

Book a strategy call

K8Lens

A read-only Kubernetes dashboard focused on diagnosis — it explains what broke and why.

The problem

The official Kubernetes Dashboard was archived in January 2026, and most replacements show what’s running rather than what’s wrong.

Engineering decisions

Diagnosis first. Plain-language diagnostics, a cluster health score, and blast-radius analysis with dependency graphs.
No agents, no Prometheus. Metrics come from Metrics Server, with a Kubelet Summary API fallback. One container image, local or in-cluster.
Read-only by design. Needs only get, list and watch RBAC. Works with kubeconfig, bearer token or service account on Kubernetes 1.24+.

Minutes

A meeting analyzer that turns recordings into decisions, owners and dates.

The problem

Decisions made in meetings vanish into recordings nobody rewatches. Follow-ups slip.

Engineering decisions

Structured extraction. LLM output is validated against a schema — every item links to a timestamp.
Queue-based pipeline. Transcription and analysis run as idempotent jobs; retries never duplicate.
Privacy by default. Audio is deleted after 24 hours. Transcripts stay in your workspace.

Costlens

Daily cloud cost attribution, down to the team and the Terraform module.

The problem

Teams see the cloud bill a month late, as one number nobody owns.

Engineering decisions

DuckDB on cost exports. Billing exports are queried in place — no warehouse to run.
Tag-based attribution. Maps spend to Terraform modules and owning teams automatically.
Simple anomaly alerts. Rolling baselines per service flag spikes the same day.