AI powered daily news - Kritrim Shabdo

Overview

A pipeline to regenerate text and audio from various online news sources to give major highlights of the day.

Details

It is getting incredibly difficult to cut through the noise of hyperbole and prime-time debates on the news channel. I wrote a Quora answer a long time back on the need to fact-check, as fake news and misinformation are always being peddled incessantly.

This led me to wonder if I could source the data from traditional print media, which seems more reliable. While it would be naive to believe that any news outlet can be unbiased, we can try to know the facts as is.

The brief outline of my idea was

1. Get data from various Indian online newspaper sources.
2. Filter noise such as entertainment, gossip, etc.
3. With the help of LLM, summarize and regenerate the text as a newsreader would.
4. Convert it to an audio file with text-to-speech service.

I do not want it to be a constant barrage of real-time news as it may not have gone through various methods of scrutiny, so I wrote a CRON schedule to trigger the above workflow once a day. I decided to write a Cloudflare worker with AI models to implement this workflow.

I am not really great at CSS, so I took the help of AI to render it with a simple yet effective style.

I have been using it for the past few days. Minor benefits include reduced screen time and saving me from doomscrolling X. Sometimes I still prefer long-form text, particularly opinions and editorials, which I have plans to convert to audio too for personal consumption.

So in conclusion, I (Your Average TeChBrO) have reinvented evening news highlights.

Deployment

The final demo is hosted here

Tech Stack

• Cloudflare workers
• Nextjs (Open next)