Applied AI studio
AI that ships.
We turn slow, manual work into AI products that run in production. In weeks, not quarters.
§01 — Focus
Find the one step AI actually changes.
Then build it, ship it, and get people using it.
- 85%less legal review time
- 9s→1spage load, +47% conversions
- <5spost-to-alert target, live media pipeline
- 100+people trained to build with AI
§02 — Services
The model is rarely the product.
We build everything around it.
AI products →
Assistants, RAG, natural-language SQL, voice and vision.
Agents & automation →
Agents, MCP servers and workflows with human sign-off.
Rapid prototypes →
A working prototype on your real data in 2–3 weeks.
Full-stack product
Next.js, TypeScript, Python, Postgres, GCP and AWS.
AI enablement →
Hands-on workshops that get your team building.
Specialist builds
Games, browser extensions, computer vision, Web3.
§03 — Work
Selected work.
Enterprise work is anonymized. More under NDA.
- 01MyColoringBook.ioAI SaaS · Next.js · Supabase · StripeAI coloring-page generator, idea to launch.↗
- 02Real-time social signalsGlobal media company · GCP · GeminiClassifies posts live and alerts editors in Slack.↗
- 03Multimodal content searchDigital publisher · Cloud Run · FirestoreA whole archive, searchable by meaning.↗
- 04Legal review assistantLaw firm · RAG · extraction85% less document review time.
- 05Commerce rebuildWordPress · PHP · Core Web Vitals9s to 1s load, 47% more conversions.
- 06Video AI & live translationVision · speech-to-speechAuto-tagged video and real-time voice translation.
- 07RUBY: RewindOriginal browser games · parallel AI agents37 retro-era games and a roguelite, built in days.↗
§04 — Process
Smallest useful version first.
Real users early. No perfect specs.
- 01Map
Find the friction and the cost.
- 02Decide
What AI changes. What stays human.
- 03Prototype
Working, on real data, in days.
- 04Prove
Measured against the old way.
- 05Ship
Evals, guardrails, deployment.
- 06Adopt
Until it’s the default way.
- Observable, not opaque.
- Humans where judgment matters.
- Model-agnostic.
- Sometimes the answer isn’t AI.
§05 — How we build
Parallel agents. One accountable team.
Coding agents build in parallel. Our engineers own every line that ships.
- Explore, build and test in parallel.
- Engineers own the architecture.
- Every change reviewed and traceable.
- Quarters of work, landed in weeks.
§06 — Engage
Three ways to start.
Every engagement starts with a 30-minute call.
Fixed scope
Prototype Sprint
2–3 weeksFind out if AI changes the workflow.
- Working prototype, real data
- Architecture and build plan
- Yours to keep
Production
Product Build
6–12 weeksBuilt, deployed and adopted.
- App, data and integrations
- Evals and guardrails
- Training and handover
Ongoing
Embedded Team
MonthlySenior AI engineers on your roadmap.
- Continuous delivery
- Enablement for your team
- 30 days’ notice
You own the code, prompts and evals. NDAs welcome.
§07 — FAQ
Questions, answered.
+−How fast can you ship?
A working prototype is usually in real users’ hands in two to three weeks, and a production first version in six to twelve. Parallel coding agents compress the timeline; our engineers stay accountable for architecture, review and quality.
+−What does it cost?
A Prototype Sprint starts at $20k, a Product Build at $60k, and an Embedded Team at $25k per month. After a 30-minute call we send a fixed-scope proposal with timeline and price.
+−Which models do you use?
Whatever fits the task: OpenAI, Anthropic Claude, Google Gemini, open-weight or self-hosted models, chosen for capability, latency, cost and privacy. Systems are built so models can be swapped later without a rewrite.
+−Can you work inside an enterprise?
Yes. SSO, private cloud on GCP and AWS, internal APIs, data governance, security and legal review, and SRE handoff. Our engineers have taken AI from experiment to production inside large enterprises.
+−Is our data safe?
Data stays in your cloud wherever possible, with least-privilege access and no training on your data. If it can’t leave your network, we run open-weight models on your infrastructure. You own all code, prompts and evals.
+−What if AI isn’t the answer?
We’ll tell you. Sometimes plain software, a better data model or a process change wins, and we’d rather say so before you spend money on the wrong thing.
+−How is this site made?
By hand: plain HTML, CSS and JavaScript, no framework, no trackers. Every mark you see is one of thousands of square pixels moved by a small WebGL engine, in typefaces we drew ourselves. The real text stays in the page for search, assistants and screen readers.
§08 — Start
Tell us about the workflow.
What’s slow, manual or expensive? We’ll tell you honestly if AI helps.
- 01Reply in one business day. From an engineer.
- 0230-minute call. Workflow, data, constraints.
- 03Fixed-scope proposal. Within a week.
Received.
An engineer will reply within one business day.