Ignition Hacks 2024

Team

  • Hemaprakash Raghu

Inspiration

Podcasts or getting information was always difficult even though we had a lot of channels and ways to get it, as the main reason was personalization and straight to the point was missing.

What it does

Here every day, a simple generation form is requested to be answered by the user, which triggers several sequences that in turn generate audio on demand using various services in the background. Here a lot more personalization is done in order achieve, to get what we need from the ocean of information. Here the audio will be multilingual of choice, geo-specific and customized interest options and even speaker voice for the podcast.

Generated Audios Example

How we built it

  • Flutter (Mobile App)
  • Data Source (News API, RSS Feeds)
  • Cloud (Supabase)
    • Storage
    • Authentication
    • Database
    • Edge Functions (Deno.js)
  • Script Generation (OpenAI, Custom LLM)
  • Audio Generation (Eleven Labs, Custom LLM)

Challenges we ran into

Script Generation using LLM:

Here script generated could be not good because of hallucinations of large LLM(s), we did get around this by fine-tuning the LLM by providing various similar scripts to get a better understanding.

Accomplishments that we're proud of

User Personalisation

Personalization is achieved by requesting the user to provide their preferences such as location, topics deep interests in a particular topic and voice of the speaker.

Audio Player Listing

A solid audio player with play and pause functionality is implemented with the Android capability. Here one can view the thumbnail, title and transcript of each audio.

Multilingual Audios

Here we focused on more than 6 languages for the same or different personalized content.

What we learned

  • Flutter Mobile Development
  • Using AI Services over rest-apis
  • Working with AWS S3 buckets

What's next for Listenade

Operating Cost

Supabase Eleven Labs, and Open AI are all services that allow us to start with a free plan and pay for what we only use. Somehow we need to migrate the entire working to a custom or in-house llm.

Built With

Share this project:

Updates