AI-native studio

We build with AI — one real venture at a time.

Omazesoft is a small, AI-leveraged studio. Each venture ships real, honest work in the open before it earns a permanent place here.

Ventures

One shared toolkit, applied to more than one problem — but only ever committed publicly once it's proven.

Live

Quant AI Lab

Applied AI for finance — agents, research automation, and quant-grade modeling. Real experiments, published in the open, honest caveats included.

Episode 01: an unsupervised model found the market's volatility regimes from price alone — no labels, no hindsight.
Visit Quant AI Lab
Live

Cohort Lab

Health data engineering — harmonizing EHR and claims data into the OMOP Common Data Model, so research teams spend their time on studies, not on cleaning data. Increasingly, agentic AI on top of that layer.

Proof: a synthetic EHR extract mapped to OMOP CDM v5.4 — source profiling, a documented mapping spec, and an enforced CI quality gate on every push (20 pass / 4 warn / 0 fail, 97.6% mapping rate).
And: a four-agent cohort-feasibility PoC — LangGraph, tools over MCP, guardrails between the agents rather than in the prompt, and a golden set that gates CI. Reproducible run to run, or the build fails.
Visit Cohort Lab
Live

Callsheet Lab

AI-native video production — ads, brand films, and recurring channel content, built with an AI-assisted pipeline at a fraction of traditional cost and turnaround.

Visit Callsheet Lab
In the lab

New ventures, incubating

Ideas get tested in private before anything gets a name here — that's deliberate. If you're building something and think there's a fit, say hello.

Get in touch
How we work

AI makes it possible for a small team to build genuinely ambitious things. It also makes it easy to spread too thin, chasing whatever a model release makes newly possible. Three rules keep that in check.

01

Proof before promotion

Nothing gets a page here until it's real work with a real result — not a plan, not a pitch.

02

The moat isn't the model

Every frontier model release commoditizes production capability a little more. What we build has to hold up on trust, data, and relationships — not on "we used AI."

03

One wedge at a time

AI multiplies focus — it doesn't replace it. We'd rather do one thing for real than three things loosely.

How we deliver

Most "AI implementation" shops either lack the domain depth or the engineering discipline to get past a proof of concept. We're built around a different split.

01

Business analysis first

The real work is understanding the client's processes and data before anything gets built. AI accelerates delivery — it doesn't replace the analysis.

02

Domain expertise, on demand

We don't claim to be the deepest experts in every vertical. When a conversation needs that authority, we bring in the right person for it — not before.

03

Engineering discipline, checkable

CI/CD, automated quality gates, and security-and-compliance-aware practice aren't slides — they're how every deliverable ships.

04

Real data, real environments

We build for production against real enterprise data sources — not one-off proofs of concept that never leave the sandbox.

Seen in practice: a synthetic-data OMOP CDM v5.4 pipeline with an enforced CI quality gate on every push — source profiling, mapping spec, and automated data-quality checks, open on GitHub ↗