Skip to content
Start a brief
Menu
Back to the gallery
  • Expertise
  • N° 0042 / Capabilities_
  • 8 disciplines
  • 41 people

What we are good at. And what we turn down.

Eight disciplines, one default stack and four house rules. Read them before you write to us.

See the eight Applied AI

01 / Capabilities

Eight disciplines, one team.

Showing 8 of 8 disciplines

  • 01 / AI

    Applied AI

    Models that sit inside a workflow, measured against a test set you can read.

    We ship
    Retrieval and agent features, evaluation harnesses, guardrails, cost dashboards.
    Typical stack
    Python, TypeScript, Postgres + pgvector, vendor-neutral model gateway
    Typical team
    2 ML engineers, 1 product engineer, 1 producer

  • 02 / Build

    Platform engineering

    The unglamorous foundation: services, queues and pipelines that stay boring at 3 a.m.

    We ship
    Service architecture, internal platforms, CI and release tooling, observability.
    Typical stack
    Go, TypeScript, Kubernetes, Terraform, OpenTelemetry
    Typical team
    3 to 5 engineers, 1 producer

  • 03 / Data

    Data and integration

    Getting the legacy system to talk to everything else without rewriting it.

    We ship
    Event backbones, warehouse models, migration plans, reconciliation tooling.
    Typical stack
    Postgres, Kafka, dbt, Airflow, plain SQL
    Typical team
    2 to 4 data engineers, 1 producer

  • 04 / Build

    Product build

    From a one-page brief to a v1 in production, with your team inside the repo from week one.

    We ship
    Web and mobile products, design systems, admin tooling, launch support.
    Typical stack
    Next.js, React Native, Node, Postgres
    Typical team
    4 to 6 across design and engineering

  • 05 / Build

    Legacy modernisation

    Strangle the old system one route at a time. Never the big-bang weekend.

    We ship
    Characterisation tests, seam maps, incremental rewrites, cut-over runbooks.
    Typical stack
    Whatever you run today, then Go or TypeScript
    Typical team
    3 to 5 engineers, 1 producer

  • 06 / Data

    Decision analytics

    Forecasts and dashboards that show their error bars and say when they are guessing.

    We ship
    Forecast services, experiment tooling, metric layers, decision logs.
    Typical stack
    Python, SQL, Postgres, a thin React front end
    Typical team
    2 data scientists, 1 engineer

  • 07 / Run

    Reliability retainer

    Two of our engineers on call alongside yours, with a pager rota and a monthly report.

    We ship
    On-call cover, incident reviews, load tests, capacity plans.
    Typical stack
    Your stack, our runbooks
    Typical team
    2 engineers on rotation

  • 08 / Run

    Technical advisory

    A senior second opinion before you sign the contract, hire the team or pick the vendor.

    We ship
    Architecture reviews, build-or-buy memos, hiring panels, diligence reports.
    Typical stack
    Reading, whiteboards, a short written verdict
    Typical team
    1 principal, 1 to 2 weeks

02 / House rules

Four rules we do not bend.

  1. Own the failure mode.

    Before we write a feature we write down how it breaks, who sees it first and what happens next.

  2. Measure before opinion.

    A claim about quality needs a test set. A claim about speed needs a number. Taste can wait its turn.

  3. Leave a smaller footprint.

    Fewer services, fewer vendors, fewer clever bits. Every dependency is a standing cost in your name.

  4. Hand over the keys.

    Your engineers merge to main in week one. By week twelve they should not need us, and we say so.

03 / Defaults

The stack we reach for, and why.

Defaults, not dogma. A brief can override any row with a written reason.
LayerDefaultWhy
LanguageTypeScript, Go, PythonThree languages cover almost every brief. We add a fourth only with a written reason.
DataPostgres firstRelational until the access pattern proves otherwise. Search and vectors live beside it, not instead.
RuntimeContainers on Kubernetes or a managed hostWe pick the smaller option your team can actually run at 3 a.m.
ModelsVendor-neutral gatewayPrompts and evals are ours to move. A provider change is a config change, never a rewrite.
DeliveryTrunk-based, flags, preview per branchSmall merges, reversible releases, and a link a non-engineer can click.

04 / The team

41 people. Mostly builders.

No account managers between you and the engineers. Producers keep the calendar honest.

  • Engineers24
  • Data and ML7
  • Designers6
  • Producers4

05 / We turn down

Saying no is a feature.

  • Staff augmentation sold by the hour
  • Fixed scope with no discovery week
  • Rebuilds nobody inside the company owns
  • Features whose failure mode nobody will write down

Tell us what has to work. We will tell you if we are the right team.

Start a brief