I'm Chris, and I created Bullet Points, an AI-assistant to help you stay briefed on any topic in the news.

You give Bullet Points a topic, and it shows you the top stories related to that topic, with AI-generated summaries and links to the underlying sources to dig further. If you like what you see, you can subscribe to daily or weekly briefs for that topic, delivered to your inbox. Whether you're a professional keeping up with industry news, a student researching a particular subject, or simply an informed citizen, Bullet Points is designed to keep you in the loop efficiently.

I've implemented this project with open-source LLMs, running on CPUs. I made this choice to minimize costs, and because I wanted to learn how to work directly with LLMs. I'm using the following models:

Instructor for embeddings (https://huggingface.co/hkunlp/instructor-large)

BART, fine-tuned on the CNN Daily Mail dataset, for summarization (https://huggingface.co/facebook/bart-large-cnn)

FLAN T-5 for curation (https://huggingface.co/google/flan-t5-base)

I'm using pgvector for vector search.

Bullet Points scans thousands of news sources, aiming to cover a broad range of topics, regardless of your location in the world or your industry. However, for now, Bullet Points is focused on English-language news.

I'd love to hear feedback from the HN community, in a comment here or a note to [email protected]. Thanks for trying it out!