Echo

AI Agent that reads the news for you. Generates a personalized podcast every morning based on your interests and what you've previously learned

Comment

acute problem

I struggle to keep up with all the news publications I'm subscribed to, and when I do actually get to reading them, much of the information is background info or fluff.

So, I built an AI agent to read the news for me.

I tell it my interests, and it reads from leading publications (TheInformation, Bloomberg, WSJ...) to create a personalized podcast briefing for my morning walk to class.

Importantly, it avoids adding any background information and focuses only on new insights/events (knows what I've previously heard about!).

The briefings are 8 minutes long, with 4 top stories. A text with a link to my briefing is sent every morning.

You can sign up or listen to a sample briefing.

bigger picture

At a higher level, there are a few activities/behaviors that I am certain will be "disrupted" in the next few years. Opening WSJ.com and reading the front page stories is one of those behaviors.

LLMs, good text to speech, and airpods have made this new concept possible. Reading articles you don't particularly care about and that have a lot of information you already know is not engaging. Gen Z reads less than any other generation and already gets much of their news on tiktok and podcasts.

An AI agent will have a good representation of your interests and knowledge. It will use this to teach/entertain you accordingly. That is what this project is an MVP of.

early validation

I tried to get some users and validation that there is a desire for this product beyond me. From some social media posts, I have +100 signups and generated personalized briefings this morning!

There were a lot of issues with SMS/iMessage sending getting rate limited during treehacks. I lost a bunch of signups and am not sure how many sadly :(. Impressions of Echo on social media posts totaled 41k during the event (though this can be a difficult measure of interest)

technical stuff

The main complexity of the project is the AI agent and podcast transcript generation. The agent is prompted as being a lower-level employee preparing a briefing for their superior. The agent is given primary topics of interest and previous stories/articles the listener has heard about. With this context, the agent can check RSS feeds from different publications, load articles, and google search topics. Ultimately the agent gathers content which is used in a RAG style approach to generate the personal podcast transcript.

The transcript generation is split into parallel and independent segments so it can quickly be generated in just a few minutes. That way we can wow the new listener as soon as possible!

On the nitty gritty, the project is React Typscript, Python FastAPI, MongoDB, OpenAI for LLMs, 11Labs for text to speech, and gcloud for hosting.

In summary

Really enjoyed building Echo and hope to keep working on it. One of my goals for treehacks was to create something I'd use personally after the event. I think I accomplished that. Take a listen to my briefing from this morning

Built With

Share this project:

Updates