Inspiration:

I'm the Director of the Seattle AI Film Festival that takes place next month at Seattle Center. We're building an agent to help our fans navigate the festival from a what point of view (ex. screenings, speakers, food and networking) and a why point of view (ex. "You're a music video fan? Okay, check out these trailers and screen times").

What it does:

The natural language agent responds to general queries, film queries, and topic queries with a conversational AI experience that aligns with the brand identity of the Seattle AI Film Festival. The agent is the front door to our brand. It functions like an app but engages the user like a character.

How we built it:

The core data source is 90 AI films from 22 countries housed on our administrative FilmFreeway account. We have full films, trailers, stills, synopses, category/genre, director bios, social media and other links for each title. The flow is extract fields from FilmFreeway to feed into a knowledge base, connect the ElevenLabs conversational AI agent to the knowledge base, design the agent against our brand guidelines, and make it embeddable across our digital touch points.

Challenges we ran into:

The biggest lift by far was creating the knowledge base for the ElevenLabs conversational AI assistant. Effectively, that took all of Saturday and into Sunday morning. I worked through the full inventory of films (>14 hours work) to extract the right information for the right fields of our core spreadsheet. Because this is a real project, I'm making XLSx first for other human collaborators and then putting that structured information into a .TXT doc for uploading to the Knowledge Base. Anybody who thinks it's write once then copy paste everywhere is going to hurt. The XLSx is for a human audience. The .TXT file is for a machine. If you don't understand that you're writing for two different audiences, you're likely to trip up. Another challenge for v2 will be the agents ability to fetch and display still image art from our films like I've attached.

Accomplishments that we're proud of:

Creating a film program for Festival Attendees isn't just about listing the screening schedule. A good film festival guide is creative media in its own right. People come to experience and to learn in that order. By building the core knowledge base this weekend, it's like we poured the foundation and framed up a house. There's still a lot of work ahead. But the bones are solid. And we're on schedule.

What we learned:

New capabilities save work in certain areas, make work in others, and force you to problem solve in different ways than what you did before. There is really NO true time savings with AI. There's a REDISTRIBUTION for where you spend your productive time. We can do so much more with AI which leads to Jevon's Paradox that the cheaper and easier it gets, the more you use it. And the more you use it, the more projects and work you take on -- typically speaking. If you're not mindful, AI can waste your time even better than social media.

What's next for Conversational Agent for the Seattle AI Film Festival:

We'll start on a legit avatar agent after the March Festival to be ready for prime time for the next event. The vast majority of conversational AI avatars suck. They're not good enough as characters to make a human connection. And their attempts to act like a character get in the way of functionality. An obnoxious character delivering a shitty experience that defaults to "get me a human !" That's the problem we're solving for our own event and for events put on by others.

Built With

  • elevenlabs
Share this project:

Updates