Bulletproof Knowledge.

By Volodymyr Pavlyshyn

What may your agent remember?
What may it claim?

An agent’s memory can preserve a mistake long after the conversation that created it. This book explores an engineering response: make accepted state changes and answers carry evidence for explicit rules.

Begin with Lean types and a small ontology. Introduce dependent types, invariants, higher-order logic, and λProlog. Build toward an agent whose symbolic checks can reject a proposed fact or answer and explain which obligation failed.

The Amazon link uses the supplied book identifier on amazon.com. Availability and formats are shown by the retailer.

Three ideas to take into your work.

1. Model knowledge precisely. Give entities, relations, and admissible states an explicit type-level contract.
2. Separate search from checking. Allow a flexible component to propose evidence while a smaller trusted component checks it.
3. State the limit honestly. A derivation is relative to the chosen definitions and axioms. It does not turn an unverified observation into a true one.

Explore the companion repository ↗

Read the description, contents, and free sample on Leanpub ↗