Bubble Up

• 

  LFG

• 

  Let's Bubble Up

• 

  So simple

Inspiration

We wanted to think big about new ways to challenge perceived boundaries of the human experience.

What it does

We identify the free-form things people are saying and aggregate the most salient aspects. Think: Conferences, Public Discussions, Employee events, Shareholder meetings, Press Releases, SOTU Addresses, Online Streaming... with big reach comes a big problem, the collective volume of the audience overwhelms the presenter, and listeners are unable to follow what their peers are thinking.

How we built it

Step 1: Collect & Pre-Process Audience Feedback The simple singular interface that allows people to participate also means that we're going to get a TON of different responses. Some comments aren't relevant, some aren't appropriate, but all of them need to be analyzed for content, sentiment, and be made available for inference.

Step 2: Aggregate the Collective Audience with the Speaker Context matters. A valid and interesting question that is not related to the topic at hand can be a distraction. The collective voice of the audience as it relates to the content needs to be considered, allowing us to determine what bubbles up.

Step 3: Reimagine Realtime Communication As the conversation changes, the relevancy of existing questions will diminish as speakers address topics of interest, the audience will be inspired to ask new questions. Conversations aren't multiple-choice questions, they are an evolution of the interaction between the presenters and those listening. No interaction is necessary to determine what the most important topic is. Speakers and audience members are always able to see what is important.

Challenges we ran into

We weren't able to connect with others as we had hoped for a full team, so we were spread thin across the stack. There's never enough time. Dependencies are stupid. Why is this still so painful.

Accomplishments that we're proud of

We made a hard thing simple, and we've turned an information firehose into a cool sip. It feels like we've cracked the code for participation from stakeholders at scale. Forget surveys, pre-defined questions, or waving your hands like a madman and hoping the MC brings you the microphone, we shoulder the heavy lifting of identifying what matters to an audience.

What we learned

Big models bring incredibly powerful solutions within reach with a small team. Good old fashioned technology matters. Orchestration across multiple technologies is valuable and big things can still be accomplished in the dark corners of areas that aren't easily automated with agentic or other straightforward approaches.

What's next for Bubble Up

We think with a few relatively minor adjustments to our approach we'll be product live and able to actively take the pulse of 10,000 audience members. We'd like to add streaming text-to-speech so we can align the incoming audience, build out dashboards for speakers and organizers to benefit from novel real-time audience data.

Built With

• amazon-web-services
• dolphin-2.9-llama3-256k
• fal.ai
• groq
• huggingface
• llama3
• snowflake