Ansibly

Inspiration

There are dozens of tools for creators, where are all the tools for the super fans?

In the age of big tech and walled gardens, Ansibly is the Content Co-Pilot for the super fan.

What it does

We are an AI engagement tool for super fans. Ansibly ingests a single podcast file and helps a fan create more content from the source. Search and ask contextual comments, generate images and memes, and drive discussions to flex your fandom. We use AI to put the crucial and interesting information you want from the podcasts you love at your fingertips in a single pane.

We incorporate the context of the media from the creator, supported by summaries, notes, and context building, to drive discussion.

In our first use case, we use the All-In Podcast as a method of showing your super-fandom.

How we built it

We used open source tools to breakdown the Podcast episode into contextual filters which are served into a companion chat to develop interesting follow up questions, additional commentary, or even visual expressions of the podcast to drive further engagement and discussion.

We utilized AI, including Daiturization and ASR, to convert media (the podcast) into rich information, thereby enabling the production of more text, images, and video content.

Specifically we developed a 3 Stage Process for enhanced Diaturization and :

1. WhisperX
2. Nvidia NeMo time alignment to reconstruct accurate text to time stamps
3. Final Pass Over Punctuation

This gives us rich information to extract into summarizations, entity extraction and POI that are helpful to the super fan.

We then use the same rich SRT file and run it through Weaviate for Podcast Search to generate vectors and contextual search to assist with prompts and more summaries to enhance content creation.

Challenges we ran into

We focused on the vision of assembling the AI powered tools into an interface that makes it easy for super fans to co-create, remix, and dive down the rabbit hole of their favorite podcasts. Keeping the demo narrow and still informative of the larger vision was certainly challenging. There was zero code before Tuesday.

Accomplishments that we're proud of

We are proud of our ability to get big pieces of the experience designed and a short demo launched on our site.

What we learned

Hybrid work is so much less productive than sitting shoulder to shoulder and hacking away.

We also learned that we would need to build our own player, and to truly sync the transcript with the speaker and other SRT files we extracted from the MP3.

We also learned that while WhisperX is faster than Whisper, it is still going to take considerable time to process the podcasts without further refinement.

What's next for Ansibly

We are going to ingest and slightly expand the offering to a total of four podcasts, including the full catalog of All-In.

Additionally, AI is only part of the experience we are developing. We have a strong believe in empowering the creators and are building on open social protocols. We have chosen Farcaster as our first social protocol and expect to stand up our Farcaster node to leverage the decentralized social graph. This will allow Ansibly to create a custom app and newsfeed of other superfans co-creating, remixing, and sharing all of the wonderful podcast insights.

Oh, and maybe raise some money.

Built With

• farcaster
• gcp
• openai
• tailwind
• vgs