Roger.ai

• 

  Architecture

Inspiration

I'm quite curious so over the years I ended up following a lot of content creators in various categories and industries. However, I quickly realized it was literally impossible to assimilate that much content, there is just not enough hours in a day. Thus came the age-old problem of the Paradox of Choice. How should I choose which content to spend precious time on when there are so many great options? That's when I first had the idea for an AI assistant that would help with the decision-making by providing concise summaries. Most existing solutions only provided article summaries, there was very little in terms of podcast or video summaries.

What it does

Roger.ai is a Messenger chatbot that summarize content from a URL, be it article, podcast or video. One of the avantages of a chatbot is that users can be onboarded without downloading a new app or creating a new account.

How I built it

Everything is hosted on Pipedream, a serverless provider that makes it very easy to connect to external APIs and provides pre-made components to reduce dev time.

The micro-service architecture (see diagram) is fairly simple, to process a message received from Messenger, the data transits from one workflow to the other. Each workflow is composed of one or more steps that run Node.js functions or bash scripts.

The AI heavy-lifting is done by the AssemblyAI API for the transcription and by the OpenAI API for the summarization.

Since everything happens on Messenger, the frontend is minimal. I used Webflow (with a template) to create something that looks good quickly.

Challenges I ran into

Until the very last minute I wasn't sure I would be available this weekend. This explains why I did not join nor create a team. I did not want to have to bail on potential teammates in case I could not make it. It was thus challenging to cover a lot of features on my own over the weekend.

Fortunately, I toyed with this content summary idea for a while and had the occasion to test the AssemblyAI API before. I started from a clean slate to write a codebase as concise and efficient as possible so that I could ship a working prototype by the end of the weekend. Using Pipedream was very helpful for that.

AssemblyAI doesn't yet provide an API to summarize text (only audio and video). So I had to use a 2nd AI provider. I only realized late that Cohere.ai had a text summary API so I went with OpenAI instead. Both AI APIs (Assembly and OpenAI) are easy to use so it was still doable quickly.

Accomplishments that I'm proud of and what I've learned

I'm particularly proud of the size of the codebase (< 500 lines). With today's tools and APIs it's possible to ship great products with minimal development.

This hackathon gave me the opportunity to discover a lot of new AI services that can unlock a world of opportunities for indie hackers everywhere. I'm really fond of the rising trend with solo entrepreneurship and I'm always impressed when I see what can be done today with less man power than just a few years ago.

What's next for Roger.ai

I have a lot of ideas for the future of Roger, among which:

• Follow content creators and automatically summarize their new posts
• A Chrome and Firefox extension to easily request a content summary from a webpage
• A WhatsApp bot for those who prefer green to blue
• A Twitter bot for aficionados of the bird
• Being able to summarize content behind a paywall or a closed community (for users with premium subscription for example)
• Automatically attend to and summarize webinars
• A Slack bot for enterprise users

Built With

• assemblyai
• bash
• node.js
• pipedream
• webflow