Data platforms and AI systems that survive production.

We build the data platforms AI needs and the agent systems that run on them. Senior engineers only, on your hours.

Engineering leadership from Booking.com and Swiss Re · Minneapolis, MN

The pilot worked. Production didn't.

In 2025, 42% of companies scrapped most of their AI initiatives, up from 17% the year before (S&P Global). The demo impressed the board. Then it met real data and a real cloud bill.

Pipelines breaking overnight? Reports nobody trusts? An AI pilot that stalled after the demo?

These are engineering problems. We fix them for a living.

AI systems & agents

AI agents that make it out of the pilot.

We build agent systems wired to your actual systems of record, route work across multiple models to control cost, and put a human checkpoint in front of every consequential action. Autonomy expands only as eval scores earn it.

Across the industry, 89% of teams building agents have observability. Only 52% have evals (LangChain, 2025). That gap is where pilots die, and it is the first thing we close.

The senior engineers who scope your system are the ones who build it. No bench, no handoff.

Data platform engineering

Data platforms have a new customer: AI.

Your pipelines were built to feed dashboards. Now they feed agents, which means reliability and cost stopped being back-office concerns. An agent reading a slow, untrusted platform ships wrong answers faster.

Our engineering leadership ran 200+ production jobs inside strict 9-hour windows at Booking.com. That is the reliability bar we build to.

What we build:

• Pipelines that finish before the workday starts
• Migrations without the multi-year rebuild
• Numbers that reconcile to source systems, monitored for drift
• Cost audits, run as a Production Readiness Sprint: if we can't find savings above the fee, it's free

Delivery teams

Senior engineering capacity, on your hours.

Hand-picked senior engineers in Brazil. São Paulo runs 2 hours ahead of Minneapolis: a full shared workday, not a narrow overlap window.

No juniors, ever. You interview and approve every engineer. They work in your standups, your Slack, your repo. No rotations without your approval.

A principal engineer leads and reviews every engagement, and answers for it. US company, US contract: you own everything we write.

Start with one engineer, or a 2-week paid pilot.

Start with two weeks

$4,900 fixed

Pick one system: the stalled AI pilot, the pipeline that fails overnight, or the warehouse bill that keeps growing. We trace it to root causes and hand you written findings with a prioritized fix list. The final call is a go/no-go, backed by numbers.

Book a 30-minute intro call

Everything is yours to execute, with us or without us. No retainer required, no pitch deck at the end.

Track record

Client work

3 engagements delivered in 2025, across the UK, the Netherlands, and Germany. One of them: a European fintech's production pipelines, processing financial data, were failing and the in-house team couldn't isolate the cause. We traced it to data-quality faults compounding with inefficient queries, fixed both without a rewrite, and added monitoring. The client maintains it today without us.

AI systems, used daily

We run our own agent systems in production. A recent competitive analysis fanned out 20 research agents and returned a decision-grade market briefing in an afternoon, every claim source-checked. The same orchestration and evals go into client systems, with cost instrumented from day one.

More case studies are published as clients agree to go on record. We would rather show two real ones than ten vague ones.

How we work

Every engagement runs on Eval-First Delivery:

Trace the failure.

Find where the system actually breaks: in the data, the code, or the model.

Build the eval harness.

Acceptance criteria become numbers we agree on in week one.

Ship behind a human checkpoint.

Someone signs off until the numbers say they don't have to.

Widen autonomy as evals pass.

The engagement ladder:

The entry point. Findings, fix list, go/no-go.

Design and build, architecture to production deployment.

Senior review of code and roadmap, from an engineer who answers.

We work in your tools: Slack, Jira, GitHub. Weekly delivery reviews. Client data never trains anything; ask for our data-handling and AI-use statement.

About Geminus

Geminus is a Minneapolis-based data and AI engineering consultancy serving US mid-market companies.

The model is deliberate: a small senior core, a vetted network of senior engineers in Brazil, and AI leverage in every engagement. No bench means you never pay for idle capacity, and nobody gets assigned to you to keep utilization up.

Geminus was founded in 2025 by Ricardo Gemignani, after a decade building data platforms at Booking.com and Swiss Re's iptiQ, with production ML shipped since 2016.

Let's talk

30 minutes, no pitch. Bring the system that's bothering you.