Dev Update: It’s a twister!

If you were looking for more technical details about boodleGPT’s development, this post is for you. I’m pleased to introduce Kisa Bostrom, Chief Data Officer at boodleAI, and Ansel Teng, Chief Technology Officer at boodleAI, who will share our progress towards bringing boodleGPT to life.

Data Team Update

Chief Data Officer of boodleAI, Kisa is affectionately referred to as “Queen Kisa”, responsible for all things data. A mechanical engineer by education, she’s built her career on growing data teams from scratch and developing useful, practical solutions. When she’s not herding cattle in her minivan, she’s raising the next generation of data artists (9, 6, 3) on a farm in Oklahoma. Connect with Kisa on LinkedIn.

If the last six months were a whirlwind in the AI space, the last few weeks have reached tornado levels here at boodle. Like everyone else, we were watching the developments happening in the AI space much like a newborn views its world: with curiosity, confusion, and a little apathy. The inventions coming out of the space were wondrous but didn’t affect us much on a day to day level.

Then we decided to jump in ourselves. We asked ourselves, how hard could it be to take a cutting edge AI technology that evolves almost daily, give it access to a database of 35 billion predictive insights, have it find the right data and answer ANY question someone could ask it, all the while ensuring ethical standards and data privacy laws are adhered to, and be ready to MVP in weeks?

The answer is: HARD.

But in less than two weeks, working around the clock, we’ve built out a layered GPT engine that connects a chat interface with our internal database, grabs the correct data, returns contextualized answers using a combination of our data and general knowledge provided by gpt-3.5-turbo, and operates in a lightweight temporary front end.

To be fair, we did have the advantage of a proprietary dataset that was ready made for layering onto a LLM and our team had previously built two applications from the ground up that are perfectly suited for incorporation into this use case.  We had a bit of a head start (okay, a lot: 35 billion predictions and 140 repositories of active code). But it’s still a daunting task. 

Now that we have the general infrastructure in place, it’s time for the hard part: testing and tuning. Right now our system has the temperament of a toddler: sometimes they are right, sometimes they are wrong, sometimes they refuse to answer, and sometimes the things they come up with are downright crazy. But that was to be expected: we have to teach and train and iterate to mature the system.  

I’ll end this update with a joke as we ride this tornado towards boodleGPT’s launch:

What did the tornado say to its annoying twin?

Sigh, clone.

Tech Team Update

Chief Technology Officer of boodleAI, Ansel is a senior technology leader since the dot com days. He has a Ph.D. in computer science and has been managing software product and system development in organizations of all sizes, from small startup to global enterprise. Connect with Ansel on LinkedIn. He loves to connect with fellow technology leaders.

Legend has it that one day in 1994, Bill Gates walked into a roomful of developers at Microsoft and announced that they should stop all current projects and focus entirely on developing a new web browser called the Internet Explorer, and the Browser War as we know it started. While it is likely to be just a myth, it certainly captured the dramatic shift at the moment.

At boodleAI, we had our IE moment recently. As the leadership team decided on the new direction of a GPT-based product development, I gathered the development team and made the announcement.  It was hard to let go of an in-development project, but with the appealing vision of a GPT-based product, the team went through the five emotional stages of change management in record time and re-oriented ourselves with the new direction.

In a true Agile manner, we quickly decomposed the product into stories and prioritized them based on what was truly needed for a beta launch. Using a car analogy, while Kisa and the data team are building the engine that interfaces with GPT, we have identified the major components that are necessary to make this a useful car: 

• The gas tank – a database that is purpose-built for boodleGPT to draw insights from;
• The steering wheel and the instrument panel – the user interface for users to provide input and receive output;
• The drive train – connecting the engine to the user interface;
• The doors – user management;
• The frame – the service infrastructure to house all of the above

We are well on our way and just about to complete all the component development and start the integration.  

As the car continues on the assembly line, we’ll provide regular updates. We welcome technical questions and/or feedback on our journey.