Beyond "Prompt and Pray"
Building Governed Agentic Systems with Geometric Memory and Verification
Four engineering books on getting agentic AI past the demo — governed agents, retrieval you can audit, testing against real ground truth, and the geometry underneath it all — and a companion for the executives who have to approve the result. Published on Leanpub.
For the engineers and architects building the system, with runnable code throughout.
Building Governed Agentic Systems with Geometric Memory and Verification
Building Trustworthy RAG with Geometric Knowledge Graphs
One relation operator built from rotation, stretch and translation
For the boards, executives and compliance leaders who have to approve the system and defend it later. No code, same architecture.
The three Beyond books cover the working system: how a governed agent is built, how retrieval earns the word evidence, and how you test the result against ground truth instead of a leaderboard. Knowledge Graph Embeddings as Geometric Operators is the mathematics they all sit on — start there if you want the foundations first, or come back to it once the applied books raise the question.
What Your AI Cannot Prove is the other side of the same table. It takes the architecture the series builds and translates it into the questions an approver has to ask, using the same running example — a governed complaint-handling assistant at a retail bank — picked up at the decision to approve it rather than the decision to build it. Read it if you sign off on these systems, or hand it to the people who do.
Every title is published on Leanpub, in PDF, EPUB and online, with free updates as the books evolve.
The books show the method. Knowlytix is the platform that runs it in production.