phidata VS NeumAI

Hi HN, I built an app to chat with arXiv papers: https://arxiv.aidev.run

I’m using function calling to interact with the arXiv api, here’s the general flow:

> For a users question, search the knowledge base (pgvector) for the topic/paper

> If knowledge base results are not relevant, search arXiv api for paper, parse it and store it in the knowledge base

> Answer questions or summarize using contents from the knowledge base.

Give it a spin at: https://arxiv.aidev.run and let me know what you think.

Its a work in progress and I’m looking for feedback on how to improve. The read time from the arXiv api is a bit slow – but not much I can do about it.

I used phidata to build this: https://github.com/phidatahq/phidata

Here’s the code if you’re interested: https://github.com/phidatahq/ai-cookbook/blob/main/arxiv_ai/assistant.py

- I used phidata to build this: https://github.com/phidatahq/phidata

Hi HN, I built an AI that can interact with the Hacker News API and answer questions about hackernews stories, whats trending, what on show etc..

Check it out here: https://hn.aidev.run

You can ask questions like:

- What on hackernews about AI?

- What on hackernews about iPhone?

- What's trending on hackernews?

- What are users showing on hackernews?

- What are users asking on hackernews?

- Summarize this story: https://news.ycombinator.com/item?id=39156778

It uses function calling to query the HN api.

To answer questions about a particular topic, it’ll search its knowledge base (a vector db that is periodically updated with the “top stories”) and get details about those stories from the API.

This is pretty barebones and I built it today in < 2 hours, so it probably won’t meet your high standards. If you give it a try, I’d love your feedback on how I can improve it.

If you’re interested, I built this using phidata: https://github.com/phidatahq/phidata

Thanks for reading and would love to hear what you think.

- Summarize this story: https://news.ycombinator.com/item?id=39156778

It uses function calling to query the HN api.

To answer questions about a particular topic, it’ll search its knowledge base (a vector db that is periodically updated with the “top stories”) and get details about those stories from the API.

This is pretty barebones and I built it today in < 2 hours, so it probably won’t meet your high standards. If you give it a try, I’d love your feedback on how I can improve it.

If you’re interested, I built this using phidata: https://github.com/phidatahq/phidata

Thanks for reading and would love to hear what you think.

Interesting to see that the semantic chunking in the tools library is a wrapper around GPT-4. Asks GPT for the python code and executes it: https://github.com/NeumTry/NeumAI/blob/main/neumai-tools/neu...