Services
What we build.
Four practices that overlap more often than they do not. Engagements are scoped and priced before work starts.
AI & Machine Learning
Automating the repetitive judgement work that currently costs you people-hours.
The hard part of an AI project is almost never the model. It is the evaluation set nobody built, the edge cases nobody enumerated, and the handoff to a team that has to trust the output. We build the harness alongside the system, so you can see whether it is actually working after we leave. This is not a recent pivot: our founder spent years at Google Cloud building generative AI proofs of concept on Vertex AI with Gemini and Claude, and running prompt engineering, model tuning, and RAG workshops for enterprise architecture teams in finance and healthcare.
Document & data extraction
Turn PDFs, scans, emails, and free text into structured records with a confidence signal and a review path for the cases the machine should not decide alone.
Retrieval & search (RAG)
Search over your own corpus that returns the right passage and cites it, with retrieval quality measured rather than assumed.
Evaluation harnesses
A repeatable test set and scoring pipeline so a prompt or model change is a measured decision, not a vibe.
Workflow automation
LLM-driven steps embedded in a real process, with the failure modes handled and a human in the loop where it matters.
Application Development
Web and internal software, built to be handed over.
We write applications the way you would want to inherit them: typed end to end, tested where tests earn their keep, documented in the repository, and deployable by someone who has never met us. No framework tourism, no clever abstractions that only the author understands.
Web applications
Customer-facing and internal apps on a modern stack, accessible and fast by default rather than as a later pass.
APIs & integrations
Typed, versioned, documented services — and the unglamorous work of making two systems that were never meant to talk agree with each other.
Internal tooling
The admin panel, the ops dashboard, the bulk-import tool. Usually the highest return per hour of anything on this page.
Data pipelines
Scheduled, observable, and idempotent — so a failed run at 3am is a retry, not an incident.
Cloud Engineering
Infrastructure sized for the load you actually have.
Cloud bills grow because architecture decisions get made once and never revisited. We size to real traffic, put the infrastructure in version control, and wire up budgets and alerting on day one — so the bill is a number you chose rather than a number you discover.
Architecture & migration
Moving what should move, leaving what should not, and being candid about the difference.
Infrastructure as code
Terraform or the platform-native equivalent, reviewed like application code.
CI/CD pipelines
Build, test, and deploy on merge, with a rollback that has been rehearsed rather than theorised.
Cost engineering
Right-sizing, scale-to-zero where it fits, and budget alerts that reach a human before the invoice does.
IT Services
The layer that keeps everything above it standing.
Small organisations tend to have exactly one person who knows how the important thing works. We reduce that number to zero by writing it down, and we make sure the backup has actually been restored from at least once.
Identity & access
Single sign-on, least privilege, and an offboarding process that actually removes access.
Backup & recovery
Backups that have been tested by restoring them, with a documented recovery time you have seen proven.
Monitoring & alerting
Alerts that fire on user-visible symptoms, tuned so that a page means something.
Technical documentation
Runbooks a competent stranger can follow, kept next to the code so they stay current.
Not sure which of these you need?
That is a normal place to start. Describe the problem in plain language and we will tell you what it would take.
Describe your problem