O'REILLY AI SUPERSTREAM: FROM DEMO TO PRODUCTION
Turning prompt chaos into reliable systems.
Turning prompt chaos into reliable systems.
Build the guardrails before the thing.
Build the guardrails before the thing.
If your team is exploring agentic analytics, you need workflows that are auditable, defensible, and ready for regulatory scrutiny. Let's talk about how our architecture can work for your data.
If your team is exploring agentic analytics, you need workflows that are auditable, defensible, and ready for regulatory scrutiny. Let's talk about how our architecture can work for your data.
Can you explain what happened on a bad day?
Can you explain what happened on a bad day?
If you cannot answer that question, your system is not ready for production. In our talk at the O'Reilly AI Superstream, we explored the gap between perfect demos and the messy reality of production deployments.
The solution is not a better prompt. The solution is a deterministic harness. By treating AI as an engineering component rather than a magic box, we built a system that separates what should happen from what is allowed to happen.
If you cannot answer that question, your system is not ready for production. In our talk at the O'Reilly AI Superstream, we explored the gap between perfect demos and the messy reality of production deployments.
The solution is not a better prompt. The solution is a deterministic harness. By treating AI as an engineering component rather than a magic box, we built a system that separates what should happen from what is allowed to happen.
Grounding AI in hard numbers.
Grounding AI in hard numbers.
When we deployed our agentic platform for financial reporting at a tier-1 insurer, we learned quickly that LLMs are optimized for plausibility, not accuracy. In regulated industries, plausible nonsense is dangerous.
Our core product, etio, solves this through a strict, four-step pipeline:
Calculate first: The system runs the math and surfaces anomalies before the AI is allowed to draft any language.
Check the math: A separate verification service confirms the numbers actually reconcile.
Critique the output: An AI Challenger reviews drafted text to ensure every claim maps back to verified calculations.
Human approval: If risk is high, the system pauses. A human must explicitly say "yes" before anything proceeds.
The result is high-integrity output: grounded in data, mathematically reconciled, and explicitly approved.
We have since broadened that approach to our other products for the financial industry and beyond.
When we deployed our agentic platform for financial reporting at a tier-1 insurer, we learned quickly that LLMs are optimized for plausibility, not accuracy. In regulated industries, plausible nonsense is dangerous.
Our core product, etio, solves this through a strict, four-step pipeline:
Calculate first: The system runs the math and surfaces anomalies before the AI is allowed to draft any language.
Check the math: A separate verification service confirms the numbers actually reconcile.
Critique the output: An AI Challenger reviews drafted text to ensure every claim maps back to verified calculations.
Human approval: If risk is high, the system pauses. A human must explicitly say "yes" before anything proceeds.
The result is high-integrity output: grounded in data, mathematically reconciled, and explicitly approved.
We have since broadened that approach to our other products for the financial industry and beyond.
OUR FLAGSHIP PRODUCT
Actuarial reporting, down to minutes.
etio
The first causal AI copilot built for actuarial teams. etio plugs into your existing systems, adapts to your standards, and delivers compliant reports in minutes — not days.
OUR FLAGSHIP PRODUCT
Actuarial reporting, down to minutes.
etio
The first causal AI copilot built for actuarial teams. etio plugs into your existing systems, adapts to your standards, and delivers compliant reports in minutes — not days.
OUR FLAGSHIP PRODUCT
Actuarial reporting, down to minutes.
etio
The first causal AI copilot built for actuarial teams. etio plugs into your existing systems, adapts to your standards, and delivers compliant reports in minutes — not days.
OUR FLAGSHIP PRODUCT
Actuarial reporting, down to minutes.
etio
The first causal AI copilot built for actuarial teams. etio plugs into your existing systems, adapts to your standards, and delivers compliant reports in minutes — not days.
Demo → Production
Bridging the gap between plausible nonsense and high-integrity output.
0
Hallucinations. Every number is mathematically reconciled before a word is written.
100%
Auditable. Every node execution and human approval gate is logged.
Already piloted with
Already piloted with
Already piloted with
The technical stack: Combining LangGraph and NanoClaw (new).
We build reliable systems by combining the right tools for the right jobs.
LangGraph is our deterministic backbone.
It is a typed state machine for agent workflows that defines the route, the transitions, and the approval gates.
Every node and state change is logged by LangGraph, making the system's behavior something you can reason about, audit, and defend.
NanoClaw enables secure, containerized execution.
It provides OS-level Docker isolation for individual agents.
By executing risky steps inside a locked container with scoped permissions, we apply the principle of least privilege to autonomous systems. The blast radius of a bad agent is contained.
We build reliable systems by combining the right tools for the right jobs.
LangGraph is our deterministic backbone.
It is a typed state machine for agent workflows that defines the route, the transitions, and the approval gates.
Every node and state change is logged by LangGraph, making the system's behavior something you can reason about, audit, and defend.
NanoClaw enables secure, containerized execution.
It provides OS-level Docker isolation for individual agents.
By executing risky steps inside a locked container with scoped permissions, we apply the principle of least privilege to autonomous systems. The blast radius of a bad agent is contained.
Review the talk slides.
Can’t find what you’re looking for? Drop us a line: contact@wangari.global.
Frequently asked questions.
What does Wangari do?
Wangari is a data analytics company that helps businesses automate their reporting and generate strategic insights from it. Create any report, hallucination-free and spending minutes instead of days. Plus board-level strategic insights that you can trust.
Which industries is this for?
We've already piloted our approach in the insurance industry. Because of our data- and standard-agnostic stance, we can work in any industry that does reporting.
What if my data is a mess?
Not a problem. We work with the data you have — our agents do the cleaning up. No data duplication or pipelining required. And if you have sensitive data, it will stay on your servers at all times —we won't even see it.
How is causation different from correlation?
Correlation tells you that things move together. Causation tells you why they move together – or why not! And we don't just tell you why, we also always provide confidence metrics, so you know what you can trust and what remains to be explored.
Can't I use ChatGPT for this?
LLMs like ChatGPT can do many useful things, but they: - Can't be trusted with sensitive data through a browser interface, - Don't remember important details, - Struggle to learn complex reporting structures, - Do not guarantee traceable output with scientific rigor, - And hallucinate if left unchecked.
Can’t find what you’re looking for? Drop us a line: contact@wangari.global.
FAQs.
What does Wangari do?
Wangari is a data analytics company that helps businesses automate their reporting and generate strategic insights from it. Create any report, hallucination-free and spending minutes instead of days. Plus board-level strategic insights that you can trust.
Which industries is this for?
We've already piloted our approach in the insurance industry. Because of our data- and standard-agnostic stance, we can work in any industry that does reporting.
What if my data is a mess?
Not a problem. We work with the data you have — our agents do the cleaning up. No data duplication or pipelining required. And if you have sensitive data, it will stay on your servers at all times —we won't even see it.
How is causation different from correlation?
Correlation tells you that things move together. Causation tells you why they move together – or why not! And we don't just tell you why, we also always provide confidence metrics, so you know what you can trust and what remains to be explored.
Can't I use ChatGPT for this?
LLMs like ChatGPT can do many useful things, but they: - Can't be trusted with sensitive data through a browser interface, - Don't remember important details, - Struggle to learn complex reporting structures, - Do not guarantee traceable output with scientific rigor, - And hallucinate if left unchecked.
Can’t find what you’re looking for? Drop us a line: contact@wangari.global.
Frequently asked questions.
What does Wangari do?
Wangari is a data analytics company that helps businesses automate their reporting and generate strategic insights from it. Create any report, hallucination-free and spending minutes instead of days. Plus board-level strategic insights that you can trust.
Which industries is this for?
We've already piloted our approach in the insurance industry. Because of our data- and standard-agnostic stance, we can work in any industry that does reporting.
What if my data is a mess?
Not a problem. We work with the data you have — our agents do the cleaning up. No data duplication or pipelining required. And if you have sensitive data, it will stay on your servers at all times —we won't even see it.
How is causation different from correlation?
Correlation tells you that things move together. Causation tells you why they move together – or why not! And we don't just tell you why, we also always provide confidence metrics, so you know what you can trust and what remains to be explored.
Can't I use ChatGPT for this?
LLMs like ChatGPT can do many useful things, but they: - Can't be trusted with sensitive data through a browser interface, - Don't remember important details, - Struggle to learn complex reporting structures, - Do not guarantee traceable output with scientific rigor, - And hallucinate if left unchecked.
WORK WITH US
WORK WITH US
Move beyond prompt chaos.
Move beyond prompt chaos.
Tell us about the AI challenge your team is facing. We'll come back with a concrete view on how Wangari can help.
We respond within two business days
You'll speak directly with our technical founders
Pilots can launch in weeks — not months
Your data never leaves your environment
Already piloted at Zurich Insurance Group
We respond within two business days
You'll speak directly with our technical founders
Pilots can launch in weeks — not months
Your data never leaves your environment
Already piloted at Zurich Insurance Group