AI & technology

Practical technology choices for systems that need to last.

We work with existing stacks first. We do not force a rebuild just to use our favourite tools.

Existing stack first

Already on one of these? We do not ask you to rewrite it.

Django
Ruby on Rails
Laravel
Go
Express.js
MongoDB

We review first. If the system can be improved safely, we improve it. Where the stack and access are clear, we usually start inside what you already have instead of forcing a rebuild.

Four principles

Every tool must earn its place. We remove tools that fail our criteria.

Four simple rules guide what we add or remove. They are not exciting on purpose. Clever choices cause surprise failures, and surprises break weekly deliveries.

Proven and widely used

We do not chase the newest framework. We pick tools used in production for years, because weekly releases depend on reliability.

One language across the stack

TypeScript from the database to the interface. That removes a whole set of bugs before they reach a release.

Everything is written down

Infrastructure is code, changes go through review, and decisions sit in a written log. Nothing important lives only in a dashboard.

You can always leave

Every tool has a way out. No vendor lock-in, and no setup that makes it hard to bring the work in-house later.

Our default stack

What we reach for on a clean-slate build.

Not a requirement, just our defaults. Every layer can be backed up, moved, or handed over to another team.

Languages

One runtime end to end.

TypeScript
Node.js
Bun

Frontend

Small, reliable, predictable tools.

React 19
Next.js / TanStack Start
Tailwind CSS
shadcn/ui

Data

Relational data you can back up and move.

PostgreSQL
Redis

Infrastructure

Runs inside your own accounts.

AWS
Cloudflare
Terraform
Docker

Quality & monitoring

Weekly releases need checks you can trust.

Playwright
Vitest
Biome / ESLint
Datadog / Grafana

Real-time, when needed

Only when live updates are part of the product.

Convex
Elixir / Phoenix

AI capability

AI that sits inside the workflow, not beside it.

We do not connect AI blindly. We decide what data it can see, what it can suggest, and what still needs human approval.

Assist

AI prepares the work so staff move faster.

LLM summaries & draftsOCR document extractionSemantic searchRAG over your documentsSupport reply draftsWorkflow assistantsReport summaries

Approve

People stay in charge of anything sensitive.

Human approval flowsSensitive actions wait for a personScoped data access

Audit

What AI saw and suggested is written down.

Audit logs for AI actionsBehind your own API keysExplainable outcomes

What we will not build on

What we refuse to build is as important as what we ship.

AI that acts without approval
High-risk AI assists staff first. It does not message customers on its own.
AI demos that cannot be maintained
If it cannot survive daily use, it was not worth building.
Rewriting what already works
A working system holds up your business. We improve it from the inside.
The newest framework of the moment
Every tool must survive twelve months of weekly releases.
No-code as the core of your product
Fine for marketing pages. Not for your main product logic.
Tools you cannot leave
If another team cannot take over within a week, we will not build on it.

Something else entirely?

Different stack from everything above? We can still ship weekly.

Share a short brief and access, and we come back with a realistic plan. We review your existing code first, and we do not force a rebuild just to add AI.

Get started

Not sure where your software project fits?

Tell us what you are trying to build, fix, take over, or improve. We will help you decide a sensible next step: a fit check, a system review, or a project discussion.

RosaanFounder, Antdragon

If we are not the right fit, we will say so early.