Inspiration
It always frustrated me to have to put in a lot of effort and research to get the answers to specific questions when I land on a site and am trying to determine if a product or service is right for me. Google seamlessly got me to the site, but then I'm on my own reading sparse FAQs. I wished there was simply a search bar that could return a fast answer, and not just run me around the site.
What it does
MagicForm is a conversational AI that is trained by the user by simply answering questions. The more questions you answer about your company, the more relevant and specific questions we generate for you to answer. After developing a sufficient knowledge base, a line of code is pasted onto the company website that renders a conversational search bar for site traffic to engage with in real time.
We focus on providing value first in the conversation by providing specific answers to questions, and then we make an offer for a conversion event once trust has been built and value has been added.
How we built it
We've been working in the conversational AI space for 3+ years deconstructing the limiting paradigms of the "chatbot" space and rearchitecting conversational flows to move away from decision trees and towards context-based understanding. We incorporate a lot of cutting edge tools that do the heavy lifting for intent generation during training as well as semantic matching during runtime.
Challenges we ran into
Building recursive knowledge generation is no easy task, and it took a lot of fine tuning of the prompts and our backend systems working together seamlessly to create a real time "wow" moment of the AI generating relevant question-answer pairs customers are likely to ask, being provided solely just other question answer pairs.
Accomplishments that we're proud of
How smooth the front end UI turned out, and how it's actually an original design. Having a conversation inside of a search bar to ask one line questions and get one line answers, no matter how the question was asked, is quite a feat in the Natural Language Understanding space. It took combining our proprietary systems we've been building for 3 years and state of the models to pull this off.
What we learned
We learned that it's incredibly important to do one thing very well, and let other intelligent and well funded teams work on the other components. Putting the ego aside allowed us to architect a system that has an incredibly smooth UX and is simply a line of code change to continue to update some of the models as others make their way into the market.
What's next for MagicForm
We are launching our beta in January for B2B SaaS companies and are very excited to start the flywheel of putting our magical search on our first handful of sites, watching it significantly reduce friction, and grow like a wildfire across the internet as more people experience it and say "hey, I want this for my site too!".
Log in or sign up for Devpost to join the conversation.