Four things, and we
say no to the rest.
Four areas, each taken end to end rather than split across vendors. Everything we build is yours from the first commit.
Original games, end to end
Design, math, art and client for real-money and social titles. We do not hand a designer's document to a math contractor and hope. The mechanic and the payout book are built together, simulated at scale, and delivered as a submission package a certifier can actually check.
- Mechanic and feature design, bonus modes, buy menus
- Python simulation, RTP and volatility targeting, max-win modelling
- Art direction, symbol and character design, Spine animation
- Svelte and PixiJS clients on the Stake Engine web SDK
- Certification-ready submission packages
AI-native products
A product where the model is load-bearing, not decorative. That means the boring half is the real work: schema-validated output, spend budgets, tenant isolation, an eval suite that runs when the prompt changes. Built multi-tenant from the first migration because retrofitting it is a rewrite.
- Multi-tenant SaaS with the isolation enforced, not assumed
- Document intake and extraction over hostile inputs
- Model output validated against a schema before it becomes a record
- Per-tenant spend budgets, asserted before send
- Audit trails an auditor would want in six months
Custom agents and agent infrastructure
Agents wired into your codebase or your workflow, with the guardrails that make the output mergeable. The gap between an agent that demos and one that works is tool design and output validation, not the model. We build the guardrails first.
- Repo agents that hold your conventions at the tool boundary
- Workflow agents on real queues, idempotent, budgeted
- Review agents, one dimension each, adversarially verified
- Hook layers, MCP servers, eval suites per agent
- Handover: your keys, your repo, no proprietary middle layer
Correctness work on systems already running
Something loses money in a way nobody can reproduce. A migration nobody will run. A feature held together by one person's memory, and that person is leaving. We diagnose first and quote the fix, not the discovery.
- Audit and teardown with a written finding, not a verbal one
- Money-path review: rounding, ordering, retries, partial failure
- Tenant-isolation and access-control audits
- Data and integration recovery
- Migration planning that survives a rollback
Diagnostic
We read the system and hand back a written finding: what is wrong, what confirms it, what it would take to fix, and what we would leave alone. Fixed fee. It stands on its own even if you never hire us again.
Build
A scoped product or system, built and shipped. Weekly demo on a real environment, not a deck. You own the repo from the first commit and can take it anywhere at any point.
Embedded
We work as part of your team on the surfaces that carry the most risk. Money paths, migrations, access control, and the agent layer. Cancel any month.
Staff augmentation by the seat
We take outcomes, not headcount slots.
Retainers to keep the lights on
If the work is done, the invoice stops.
A proprietary layer you cannot remove
Your repo, your accounts, your infrastructure.
Pilots that were never meant to ship
We build production software, not a proof of concept.
Not sure which
one it is?
Describe the problem and Monet will tell you whether it is a diagnostic, a build, or an embedded engagement. It will say diagnostic when that is the honest answer.
Describe what you're dealing with and I'll tell you which engagement shape fits, including when the answer is a short diagnostic rather than a build.