Case Studies

A sample of our portfolio. See how we can help you.

Reaper Actual

Game Execution

Reaper Actual is an open-world persistent shooter. Michael led a small dev team that shipped 10 major features alongside major improvements to reliability and framerate post-launch. Three AI developments supported the work. Project Brain gives agents the project's rules, systems documentation, and a code graph of how the systems connect, so they read how the game works before they change it. Asset Catalog is an agent-maintained database that tags the project's meshes by what they are and how they can be used, such as cover, loot surface, or structure, so agents reason about a place instead of raw geometry. The Virtual Designer uses both to do world and combat design work in the editor, placing loot and NPC encounters, and it shows the team a plan to approve before anything changes.

Visit reaperactual.com →

Usage-Based License Server

SaaS Commerce built for CinePipeAi

A licensing and usage service with a pluggable payment gateway and Python client library. It puts feature entitlements and credit usage behind one interface, so a product asks the same questions whether a job runs in the cloud or on the customer's own machine. Verification and validation are formal. A risk-to-damage register scores each failure scenario and traces it to its tests: unit and integration tests for verification, and end-to-end tests against real services for validation. Coverage tests check all critical failure scenarios. Key security decisions, including tenant isolation, are tested using Modified Condition/Decision Coverage. Over 330 automated tests verify its integrity.

Private repo

Wayfinder

AmplifiedByAI product

Wayfinder is our own project management system, built around a dependency graph instead of a sprint board. Work links to goals and KPIs, and a ready queue recomputes as things change, so the plan follows what is unblocked and what matters most. Each task carries a certainty score built from its open questions and dated confidence reports, which brings unclear work to the surface early. It is in active development and runs our own projects.

Whitepaper coming soon

Open Source

These are systems and utilities you can use in your personal or commercial development today.

Trusted Autonomy

Open source

Trusted Autonomy runs any AI agent, including Claude Code, Codex, and Ollama, on a staged copy of a project. A supervisor agent reviews every change against the stated goal and the product constitution, then routes anything consequential to a person for approval before it reaches the real system, and every step goes into an audit trail. It is written in Rust and released under the Apache-2.0 license.

View on GitHub →

Meridian

Open source

Meridian shows what agent work and token spend are buying. It classifies agent work by business category, scores each piece against the company's KPIs, and reports token spend by category, so cost can be read next to results. It runs as an MCP tool, as a standalone command line, or inside Trusted Autonomy.

View on GitHub →

Autoreward

Open source

Autoreward turns subjective quality judgments into signals a pipeline can use. It sorts each check into Measured, Predicted, or Reviewed tiers, ships a library of gauges to start from and a model index for choosing proxies for human review, and includes a multi-objective optimizer so one score cannot rise by quietly pushing another down. It supports reward-based training loops such as RLAIF and automated validation, and it is MIT licensed.

View on GitHub →

Task Graph

Open source

Task Graph plans the order of work. Given tasks and their dependencies, it works out which can run in parallel right now, grouped into waves with conflicting tasks kept apart. It is deterministic with no model in the loop, so the same graph always produces the same plan. It was extracted from Trusted Autonomy as a small standalone library, Wayfinder's scheduler depends on it, and it is released under the Apache-2.0 license.

View on GitHub →

Let's talk about your team

A first conversation costs nothing but time: what you're trying to change, and whether we're the right fit to help.

Book a call