We're currently working with Spencer Thompson in creating a GPT-powered mental health app. GPT integration is critical to our logic and UX, but it became clear quickly with extensive agent testing that it was prone to AI hallucinations – a known tendency of GPT models to respond outside the bounds of reality or the provided user context. We optimized the use of GPT agents without sacrificing UX.
We focused on what GPT models are good at: textual extraction and summarization. We created agents that focused on performing specific tasks, instead of hardcoding rules into just one agent, to prevent confusion and AI hallucination.
Instead of instructing the chatbot to “act like a therapist”, we provided specific instructions around what the AI should ask or check for from the user based on user flows designed by Spencer. This ensured that the AI would not act as a therapist but rather act as a facilitator between Spencer's content and the user's intentions.
Current GPT models have context limits that make it difficult to sustain conversations over long periods of time. To prevent this, we store relevant contextual data in a database and use helper agents to help remember things for the user.
You’ll be collaborating directly with us, Tyra and Jacoby. Together we have 10 years of combined experience owning and leading projects through end-to-end product cycles, as well as leading development, content and design teams.
Individually, we’ve grown and managed businesses and led a company through a critical ERP adoption for VC funding, so we have first-hand experience on building digital products in the broader context of running a business. You can view a little bit more about our previous involvements and projects here.
That depends entirely on your requirements. It can take anywhere from a week for simple integrations, up to 2-3 months for customized solutions like chatbots that are trained on your business knowledge. The turnaround time varies depending on: whether or not the AI model is going to be trained, what data it’s going to be trained on, and the breadth and depth of data it’s going to be trained on.
Requirements drafting is a crucial first step to our process to keep the development timeline somewhat predictable. It’ll be easier for us to provide a turnaround time if you had a list of features, functionalities, or screens you’d like the project to have.
AI integration starts at $500, on top of monthly API costs. We’ll be using your own API key which means that monthly API costs will be associated with your account.
At the very least, we would prefer if you had screens/pages, features, or a user flow in mind.
If your project is in its early stages, we recommend checking out the 360 Sprint to see if that’s a better fit for you than investing upfront on developing a product.