Services

Find your problem. We'll show you how we solve it.

We organise the firm into practices, by the kind of problem, and sell the work in four engagement shapes, by how much commitment you want to make up front.

Practices Ways to engage Start from the problem

Practices

The problems we are hired to fix, by practice.

Data Platforms

The data platform bill climbs every quarter, and pipelines break whenever a source system changes.

Apache IcebergSparkEMRGlueLake FormationSnowflakedbtStep Functions

We design and build AWS-native data lakes and lakehouses on S3, Apache Iceberg, Spark, EMR, Glue and Lake Formation, with Snowflake where it fits. Ingestion absorbs schema drift on its own, orchestration runs on events instead of an always-on scheduler, and compute scales to real usage.

Proof. University of Wisconsin: a consolidated data lake on EMR and Iceberg, with an auto-relationalization engine that turns Workday and PeopleSoft exports into queryable tables. Led by Robin Tanner.

Embedded AI & Data Architecture Data Layer Health Assessment

Field briefs: the lean data lake · auto-relationalization · event-driven orchestration · geospatial lookup

Applied AI & Agents

The AI demos work. The data layer underneath them will not survive production.

Amazon BedrockClaudeMulti-agent designRetrievalText-to-SQLAthena

We build agentic and retrieval systems on Amazon Bedrock and Claude, over your own data: schema-aware retrieval, natural-language-to-SQL, classification agents and operational triage. We start with the data layer, because that is where production AI fails first.

Proof. Salesforce: Bedrock agents that classify PII across the data lake at 95%+ accuracy, and an LLM triage system that diagnoses pipeline failures. Led by Robin Tanner.

Embedded AI & Data Architecture

Field briefs: auto-relationalization · the lean data lake

Cloud Architecture & Modernization

A migration or modernization is under way, and nobody owns the target architecture.

AWSGCPAzureTerraformMicroservicesSnowflake security

We set target-state architecture on AWS, GCP and Azure, and carry it through migration: on-premises analytics moved to the cloud, microservices, network and security design, and Snowflake security frameworks. Decisions are written down so the next team can follow them.

Proof. Viasat: the target architecture for sub-second microservices moving from AWS to GCP. Taco Bell: the company's first ground-up AWS analytics environment. Led by Robin Tanner.

Embedded Architecture Leadership Architecture Review

Field briefs: event-driven orchestration

Delivery Engineering

The release process has become the bottleneck for every team that ships.

CI/CDMerge queuesTerraformGitLabJenkinsTrunk-based development

We redesign how code reaches production: shared-sandbox integration with per-feature promotion, branch-per-environment CI/CD, infrastructure as code on every merge, and agentic tooling where it pays for itself, such as assisted conflict resolution and generated release notes.

Proof. Viasat: a selective-promotion CI/CD model that lets dozens of engineers test together in one sandbox and still ship any feature on its own. Led by Robin Tanner.

Selective Promotion

Field briefs: selective promotion · branch-per-environment CI/CD

Data Governance & Analytics Operations

Self-serve analytics grew faster than the rules for who can see what.

Row-level securityLake FormationSnowflake rolesIdentity & accessData qualitySix Sigma

Governed self-serve analytics, row- and column-level security, identity and access, data quality, and the program management that keeps an analytics operation running.

Proof. Pratt & Whitney: a governed hub-and-spoke self-serve model sustaining 300+ business applications for 6,000 consumers worldwide. Led by Magdalena Vudoyra.

Architecture Review

Ways to engage

Commit as much as the evidence supports, and no more.

Each engagement is scoped in writing. Pricing is by inquiry and set per engagement.

Most common first step

Data Layer Health Assessment

Length: 2 weeks · fixed scope
Best when: You suspect the data layer costs too much but cannot prove it
Who works on it: A principal architect, with analysis support
You get: An 11-page executive Briefing, a three-year cost model, and every quantitative artifact
Commitment: Fixed scope. No vendor commitment attached

See the assessment

Architecture Review

Length: 2-week review window
Best when: One decision has to be defensible before you sign
Who works on it: A principal architect
You get: A written review with findings, risks and a recommendation, plus a walkthrough
Commitment: Fixed scope. Optional 60-day follow-up

See the review

Embedded Architecture Leadership

Length: 1–2 days a week · quarterly
Best when: Your engineers are strong but nobody holds the architecture
Who works on it: A principal, embedded with your engineering leads
You get: Decision records, reference architectures and a quarterly executive briefing
Commitment: Quarterly contract, reviewed each quarter

See embedded leadership

Architecture-Led Delivery Team

Length: Scoped to the build
Best when: The build needs more hands than your team has
Who works on it: A principal lead, architects, engineers, program management and partner capacity
You get: A working platform, runbooks and a structured handover
Commitment: Scoped in writing, reviewed each quarter

See how teams are staffed

Also available as a focused engagement: Selective Promotion, for engineering organisations whose release process has become the bottleneck.

Start from the problem

Fifteen situations that usually lead to the first call.

Each one routes to the engagement that typically fits it. If yours is missing, describe it to us anyway.

01 / 05

The bill

The data layer costs more than it should, and the number is hard to defend.

The data platform bill climbs every quarter, and no one can say which workloads are driving it. Data Layer Health Assessment A multi-year Snowflake or Databricks renewal is coming, and there is no evidence base for whether the commitment is sized right. Data Layer Health Assessment Data costs cannot be allocated back to the teams and products that generate them. Data Layer Health Assessment

02 / 05

The decision

A specific call has to be made, and it needs a defensible, independent answer.

A vendor has proposed an architecture, and the organization wants an independent read before it signs. Architecture Review Two internal teams are advocating different approaches, and leadership needs a defensible tiebreaker. Architecture Review A data platform was inherited, and no one can say with confidence what is load-bearing and what is safe to change. Architecture Review

03 / 05

The platform that pages you

It works, mostly — until it doesn't, and the failure modes have become routine.

Upstream schema changes keep breaking pipelines, and the patches have accumulated into a structure no one fully understands. Embedded AI & Data Architecture Every new data source takes weeks to onboard when it should take days. Embedded AI & Data Architecture Queries have slowed to a crawl, and the cause is somewhere in a million small files no one has had time to compact. Embedded AI & Data Architecture

04 / 05

The AI that won't survive production

The demos land. Production is another matter — and the data layer underneath is the reason.

The agent and retrieval demos work, but the data layer underneath them will not hold up in production. Embedded AI & Data Architecture An LLM is being wired into an operational workflow, and the patterns are being set without a designer in the room. Embedded AI & Data Architecture The board has asked for an AI roadmap, and there is no internal author for it. Embedded AI & Data Architecture

05 / 05

The missing architect

The team is capable, but no one is holding the architecture — and it shows.

The engineers are strong, but there is no architect, and decisions are accumulating faster than the team can absorb them. Embedded Architecture Leadership The organization is between architecture hires and needs a steady hand for six to twelve months. Embedded Architecture Leadership Technical posture is invisible to the executive team — the board hears nothing it can act on. Embedded Architecture Leadership

Platforms

Your stack is probably on this list.

Platforms our team has built on in production. These are not partnerships or certifications.

AWSAmazon BedrockClaudeApache IcebergApache SparkSnowflakedbtGoogle CloudMicrosoft AzureTerraformPostgreSQL / PostGISPalantir Foundry

Bring us the problem you can already name.

The first conversation is free. If we are the right firm for it, we propose a scope in writing. If we are not, we tell you who to call.

Book a first conversation Start with a 2-week assessment