What is Bluejay's purpose?

Welcome back. I took time this week to consider Bluejay's purpose. What is Bluejay? What is our charter? I guess we get the luxury to ponder these questions now that Rohan and I aren't a team of two anymore :)

Of course, Bluejay is a testing and monitoring platform for Conversational AI agents. We run life-like simulations in pre-production environments by creating digital humans – (yes I write with em dashes, though ChatGPT ruined their elegance) synthetic replicas of your customers, and we monitor every conversation your agent has with real customers to flag errors.

But is that our purpose?

Every day, we read headlines about flagship models advancing in capability and scope of access. We hear of "sentient" Anthropic models being aware that they are being tested. We see AI-enabled crime, from impersonation to voice cloning to illicit image generation. We even hear of AI agents being used to potentially conduct autonomous military operations without human oversight, or be used for mass surveillance without opt-out.

Does AI act beyond human governance? Does it amplify bad actors? Is it sentient? All of these questions point towards one truth:

Human trust in AI is at an all-time low.

To increase human productivity and usher in a golden era where human and machine brains can build the foundations of our future, humans need to trust AI – but we can't trust what we don't understand.

And that brings us to why we are really building Bluejay. Why we are diligently simulating every interaction, monitoring every conversation, and flagging every error:

We will provide the tools for humans to understand, improve, and oversee artifical intelligence. With these tools, we will usher in a golden era of trust and productivity that society will reap the benefits of for millenia.

Announcements

Here's what happened at Bluejay last week:

• Faraz spoke about building robust evaluation frameworks for voice AI agents at the Agora Voice AI Mixer.
• Rohan flew on-site with our customer's customer to learn more about their use-case.
• Dev, our intern, released technical research article about Voice AI testing for full duplex (speech-to-speech) models – an absolutely essential piece to read.
• Last week, the team pushed 34,639 lines of code to make Conversational AI reliable.

Feature Spotlight: Threshold Alarms

Introducing Threshold Alarms! Now, users can be notified if their agent runs into repeated failures over a specified time window.

For example, users can now configure an alarm to be notified if their agent's average latency is over 4000 milliseconds in 5 conversations over the past 10 minutes. This could be an early warning sign of a node being overwhelmed – something that Bluejay can now help you discern.

Coming Soon

• Upcoming Events: We're joining forces with two very popular voice AI companies for an exciting event this month. More on this soon!
• Upcoming Features: This week, we will add analysis features to our simulations to help users gain actionable insights as to what they need to change in their next simulation run!

That's all for this week. I'll see you next time!

Faraz Siddiqi
Co-Founder & CTO @ Bluejay