Mary Maker: AI Wedding Planner

• 

  mary maker

• 

  Example conversation snippet 1

• 

  Example conversation snippet 2

Inspiration

Hannah and I are not just partners in this hackathon, but also are in the process of planning our wedding. The research, and then the back-and-forths with vendors are time-consuming and introduce a lot of decision anxiety. Since we're too cheap to hire a wedding planner, we tried to create one ourselves!

What it does

Mary Maker has a conversation with someone and tries to understand what kind of wedding they're looking for. From there, she collects necessary information to contact vendors. She contacts them via email while CCing the client(s). She can also provide daily or weekly email updates leading up to the wedding, providing recommendations for catering, music, ceremonies, etc.

How we built it

We used Voiceflow for the conversation flow, most GPT-4 connections, and Web Messaging interface. We used API integration steps to connect to our Python server to handle web searching, some GPT-4 connections, and emailing.

Challenges we ran into

Decisions on how to design the Conversation flow were a bit challenging. How open ended to make it, how dynamic to make the prompts. As well, we didn't hard-code any of Mary's responses, and totally used GPT-4. That made controlling the conversation flow a bit challenging, but makes the conversation nicely flexible.

Accomplishments that we're proud of

Getting a working prototype in a day of coding was really awesome for us as we hadn't worked as a team before. Building a pipeline for GPT-4 to generate ideal search terms specific to the couple based on the conversation, and then building a custom web scraper to get results to feed into the next prompt was really cool to see working.

What we learned

Too much about merging git conflicts and sending emails programmatically. A bit more about prompt chaining and web scraping.

What's next for Mary Maker: AI Wedding Planner

Better, more diverse prompts. We want to subtly update the prompt (especially the header before the conov) and run also parallel LLM checks for things like entity detection and overall conversation goal. We want to add a database for longer-term planning, hook up the Pinterest and Google Sheets integrations, and dial in our email scraping and email prompting.

Built With

• flask
• gpt
• python
• voiceflow