Skip to content
01For your CTO

Everything runs in your cloud account, and stays there.

This page is for whoever looks after your systems, so it gets into the technical detail the rest of the site leaves out.

Your cloud account
Karasu works in here, with access you grant and can revoke.
  1. AIPrompts, evaluations, outputs
  2. Pipelinesdbt™: validation, reconciliation, history
  3. DataApache Iceberg tables, in your storage
  • Snowflake
  • Databricks
  • Microsoft Fabric
  • DuckDB
All four read the same tables directly. Nothing is copied out.
Held by KarasuNothing.
02

Data

We store data as open tables in your own cloud account, with nothing proprietary in between.

FormatApache Iceberg, readable by Snowflake, Databricks, Microsoft Fabric or DuckDB.
LocationYour cloud account and your storage, under your access controls.
Transformationsdbt models in your own repository, version controlled and tested on every change.
If we part waysThere's nothing to migrate, because it was always in your account.
03

Stack

SnowPro Core certified.

We build on what you already run. These are the tools we know well, grouped by the layer they serve.

  • Snowflake
  • Databricks
  • Apache Spark
  • dbt
  • Airflow
  • Prefect
  • Python
  • AWS
  • Terraform
  • Docker
  • GitHub Actions
Trusted dataSnowflake · Databricks / Delta Lake · dbt · Airflow · Prefect · Snowflake Streams + Tasks · Dynamic TablesPlus Snowpipe Streaming and Debezium for change data capture.
Data qualitydbt tests (dbt-expectations) · Great Expectations · ElementaryChecks run on every load, with alerts to Slack.
Enriched dataPython · Snowpark · PySpark · Entity resolution and fuzzy matchingMatching the same insured across systems before anything is joined.
Models in productionAmazon SageMaker · Containerised model endpoints · Retraining on Snowflake TasksFive models taken into production this way.
AI on your documentsClaude API · Snowflake Cortex · Document AI · MCP serversRun inside your account, against models you approve.
InfrastructureAWS (S3, ECS / Fargate, EKS) · Azure (ADLS, Data Factory) · Terraform · Docker · GitHub ActionsEverything as code, deployed through tests before it reaches production.
GovernanceRow access and masking policies · Tags · CIS Snowflake Benchmark reviewsFive accounts reviewed against the CIS benchmark to date.

On another stack? Most of this has a close equivalent elsewhere, and the data map will say plainly if your setup is outside what we know.

04

AI

AI only goes in once the data underneath it is reliable, and it runs under the same controls as the rest of your systems.

Where it runsIn your own environment. Data stays inside the boundaries you set.
What you ownPrompts, evaluations, pipelines and outputs, stored alongside the data.
ModelsWhichever providers you approve: [list]. Karasu has no model of its own to sell you.
How it's judgedAgainst the baseline, on hours saved and error rates. We don't report an accuracy score without something to compare it to.
05

Continuity

Karasu is a one-person firm, so you'll want to know what happens if Alex is unavailable.

Small releasesEach piece is finished and working before the next one starts, so a pause never leaves you with something half-built.
HandoverWe write the handover notes as we go, so your team could take over at any point.
Everything in your cloudWe don't keep anything on our side that you'd need back.
InsuranceProfessional indemnity: [insurer, cover].
Engagement letterIt writes down the baseline, how we'll measure it, and that you own what we build.

dbt and the dbt logo are trademarks of dbt Labs, LLC. Karasu Intelligence is not affiliated with dbt Labs or with any other vendor named on this page.

More detail for your team?

Any build after the data map comes with a technical proposal your team can go through line by line.

Book a data map