Ben Taylor
CTO and co-founder, Rainbird
Has spent more than a decade building deterministic reasoning engines. Shows the architecture, the engine and the evidence tree.
LinkedInLive webinar and demo
Your agent shouldn't be making critical decisions
Learn how to build a model your agents can call for decisions that are precise, deterministic and auditable, in milliseconds.
Thursday, 3 December 202615:00 GMT · 10:00 ET45 minutesOnline · Free
Register free Can't make it live? Register anyway and we'll send you the recording.
Who this is for
You're building AI agents, and some of the decisions they make have to be right every time. You might be a developer, a technical founder or part of an AI engineering team. You might be the person who has just been told the agent can't ship until someone can audit what it decides. Either way, this session is technical, practical and built around a live demo. Bring your questions and your hardest use case.
What you'll see in 45 minutes
A short demo that shows what deterministic reasoning looks like in practice.
The problems we kept running into when building agents, and why we built Rae.
Rae end to end: from a brief and a set of source documents to a tested model with a live endpoint.
Connecting that model to an agent over the MCP endpoint, and to an application over the REST API.
Reading an evidence tree: what it shows, and how it changes the way you debug.
How to start on the Rae free tier, then your questions. Technical ones first.
You've probably hit this wall
The demo works. Then real data arrives and the agent starts giving different answers to the same question, and when it gets one wrong, nobody can say why. For most features you can live with that. For decisions that have to be right every time, you can't, so the project stalls.
The problems stack up:
A 200-page policy is reinterpreted every time a language model reads it as context. Change nothing and the answer still shifts, so you can't unit test your way to confidence.
A reasoning trace is a simulation of reasoning, not a record of it. Debugging means reading traces and guessing, and "the model said so" won't satisfy a reviewer or an auditor.
Guardrails rest on a risky premise: let the model make the mistake, then hope you catch it. Using one LLM to judge another adds a second non-deterministic layer on top of the first.
Agents talk to tools in prose. Formatting slips, instructions get half-followed, and every call is another place for drift.
A human reviewer feels like a safety net, but people under volume and time pressure learn to trust confident outputs. The research on automation bias is not reassuring.
We kept running into this ourselves, so we built Rae.
What Rae does
Rae, the decision automation platform from Rainbird, turns regulations, policies and your experts' knowledge into a model your code or your AI agents can call over the MCP endpoint or a REST API. There's no language model in the decision path. The same inputs give the same answer every time, with an evidence tree that shows why.
Language models still do the building. You write a short brief and add the documents. Rae researches them, proposes a plan, builds the model in Raelog, Rainbird's knowledge representation language, then tests it and tries to break it. Every decision comes from Raelog Reasoner, a deterministic inference engine that combines the model with the facts of the case in milliseconds.
In practice that means:
The same facts and model version always return the same decision and evidence, so your tests mean something and regressions get caught.
Every decision returns an evidence tree showing the data and rules behind it. Debugging becomes inspection, not archaeology.
Any agent, application or chatbot can call the model through the MCP endpoint or the REST API.
Run it headless when every fact is supplied, or let it ask for what's missing instead of guessing.
The model is yours: readable, versioned, and shipped with its test cases and documentation.
When policy changes, add the new document; the same tools amend, test and deliver an updated model under version control.
What you get when you register
Join us live, and we'll send you the recording afterwards.
A direct line to the people who built Rae. We keep the last ten minutes for your questions, technical ones first.
A head start on the Rae free tier. Rae is in open beta now, so you can try what you've seen the same day.
Who's presenting
Lucie Hunt
VP Enablement, Rainbird
Runs the live build, from brief to tested model to endpoint, and takes the how-do-I questions.
LinkedInLive webinar and demo
Bring your hardest use case
Rae is in open beta and we want developers to put it to the test. Register, bring the decision your agent can't be trusted with yet, and ask us the toughest questions you have. We'll show you how to make it precise, deterministic and auditable.
Register free