What we build

AI Readiness Scorecard

A self-serve version of the audit we run with clients. Answer a few questions and get a straight read on where you stand, in plain language.

• A straight score on your data, tools, and team
• The two or three things worth doing first
• An honest note on where AI will not help you

AI Engineering Cookbook

The playbook our own engineers follow when they build software with AI agents. Agents write most of the code. Humans set the intent up front and check the result before anything ships.

Internal tool. We built it for our own team, use it on every project, and published it in the open.

• Plan before you build, so the agent knows exactly what to make
• Checks and guardrails, so nothing ships without being verified
• Works on new projects and on codebases you already have
• Installable skills and templates your team can use today

Recursive Agentic Improvements

The tooling our own engineers use to build AI agents and keep improving them. It scaffolds a new agent, tests how it actually behaves, then refines it step by step instead of guessing.

Internal tool. We built it for our own team, use it on every project, and published it in the open.

• Three commands that create, improve, and extend an AI agent
• Works across the main agent frameworks: Agno, CrewAI, LangGraph, and Google ADK
• Reads the live documentation first, then writes a plan you approve before any file changes
• Tests run offline against mocked models, so the results are repeatable

Hybrid Search RAG

A working example of the search layer we build into AI products. Most AI assistants are only as good as what they can find. This one runs two searches over your data at once — one that understands what the question means, and one that matches the exact words — then merges the results so the best answer comes out on top. A recipe recommender is used to keep the example easy to follow.

• Finds the right answer even when the customer does not use your exact wording
• Still respects exact terms like product codes, part numbers, and names
• Runs on your own database, so answers come from your data, not the open internet
• Built with tools you can run yourself: CrewAI, Qdrant, Neon Postgres, and Google Gemini