Safety

Yanzio has learned how to use AI to get relevant, factual, repeatable, accurate and credible results at scale, and in ways that make data usage and manipulation by traditional software and databases extremely powerful. And all this is specifically  for I&R data.

We methodically tune, test, iterate and perfect the usage of AI for each specific task at hand and, once it is working well, we scale it up to work on large datasets. To date we have performed many millions of API requests in this manner specifically on multiple I&R's data from their resource databases. A simple example of a task is: “Apply a specific I&R Style Guide rules for Eligibility on every single Program record in that organization's database.”

Yanzio currently uses a variety of AI Language Models (LMs) and each tend to excel at different tasks. New ones are being released very frequently and we're constantly evaluating them for a wide range of tasks. If you have requirements for data security or sovereignty and require a current list, please contact us.

By accessing these LMs directly via API, we have far more control than a casual user of ChatGPT of "telling" the AI what we want it to do and how we want it to structure its responses.

For the traditional software components like databases, web application code and networking, Yanzio exclusively uses the Microsoft Azure cloud system and all of our software code, storage and networking systems are in their data centers. 

For AI LMs, our vendors are in various different data centers.

So when we run our AI processes, your data travels to our AI vendors. The data is sent in small chunks usually no bigger than one field value (like Description or Eligibility) from one single record, at a time. 

When we add AI vendors, we scrutinize their security and data policies to ensure we can fulfill our legal commitment in our Software License Agreement with you to protect your data and your full data ownership rights.

Your data may also travel to our other service vendors who supply routine services like security, website monitoring and email communication services.

Absolutely not. Yanzio makes it clear in our Software License Agreement that you alone maintain ownership of your data. We will only work with vendors who fully support that right.

Neither Yanzio nor our AI vendors will modify the AI or the LMs themselves with your data. 

Yanzio may use your feedback and suggestions and other forms of input to improve the way we use AI and non-AI systems and tools, as described in our Software License Agreement.

We employ a variety of methods to limit episodes where the AI embellishes or "hallucinates". This includes finding the best LM for each particular task, like when we are assessing an Eligibility answer from your database to evaluate its adherance to your Style Guide rules.

We also use other LMs to check the initial work done by the first LM on a task. This may sound odd, but some LMs work really well in this capacity of checking other LMs, even if they aren't great at generating the original evaluation.

And there are strong and emerging tools and approaches in the AI world, like the "RAG Triad" and "Observability" that allow us to inject fidelity, accuracy and quality assurance into our AI processes.

Does this mean that AI will always give a perfectly accurate answer? No, but we continue to see very high accuracy rates and we are improving that nearly every week.

That's also why we present the AI's evaluation, score and recommended changes to you. Your expert eyes can review its work before you consider making any changes to your data.

Yanzio's founder started and built one of the major I&R platforms still widely used today, iCarol. Since it also served crisis centers and other mental health entities, it is required to be HIPAA compliant.

We are taking the same approach to securing your data, using the best tools from Mircrosoft and third parties to protect and audit all of our systems and your data within them.