Daan 高大安

(🧵6/6) I enjoy building (random) side-projects like this and am always open to feedback on my projects, so let me know what you think! Check it out: flows.news

(🧵5/6) Fully built within @vuejs and @nuxt_js, and fully hosted on @CloudflareDev using Workers, D1, R2, Workers AI + Gateway and Workflows. To keep it running all day, seems to be about 125k tokens in and 220k out for the AI use. Not sure about the cost yet.

(🧵1/6) I like to read the news and its part of my daily routine to read several different news sources a few times a day. Until someday, I was thinking, why do we read all news chronologically? Wouldn't it make more sense to read it as causation and results?

(🧵6/6) Let me know if it helps you out or what I can improve on to make my boilerplate more useful! Check it out here: github.com/daandegroot123…

(🧵5/6) Having both front-end and back-end in one codebase has helped me in particular, since it easier to provide context to my coding agents. I also connected Cloudflare, Nuxt and Nuxt UI MCP servers (in VS Code) to help out.

(🧵1/6) My go-to stack to build my web apps with is only using @nuxt_js + @CloudflareDev. Full-stack and easily deployed to Cloudflare Workers, where I pay $5/month for everything I need (D1, R2, AI, vectors, logging, gateway, workflows, etc.)

@SanderVgc @RakeyVGC when using dragonite?

I made day 2 and got top 32 at the Stockholm Regional Championships! SO @RakeyVGC for helping me with this amazing team and the Dutch homies for making my first Day 2 unforgettable❤️

@CShorten30 Aha makes sense, thanks for the explainer!

@daandegroot69 @devorein @weaviate_io Hey @daandegroot69, both are counting tokens it’s just split with “prompt” and “completion” for the billing interface — definitely makes a bit more sense for the generative apis than embeddings. 550k / 4.2k ~= 130 tokens on average per embedded document which makes sense

(🧵1/7) This weekend I started exploring vector databases, following @fireship_dev's video, so I created a @weaviate_io database filled with all @ycombinator companies! That allowed me to create this fun visualisation of the relatedness of companies throughout all the batches. 🚀

@weaviate_io It is set; that's why it's weird. It also worked on my laptop earlier but not now...

@daandegroot69 Yeah - the query is correct but the content isn't vectorized. nearText only is available if the class (in this case Company) has a vectorizer set.

@weaviate_io But not this error, right? Just copied the code from the docs and put my own class name in

@daandegroot69 It needs a vector to do the search :) But it will present an error if it can't generate a vector

@weaviate_io My OpenAI usage was mainly for the initial text2vec to populate my data, not for the searches (although as I am testing now, on my PC compared to my laptop over the weekend, the whole option of 'nearText' doesn't seem to work 🤔)

@daandegroot69 @devorein @fireship_dev @ycombinator Are you using hybrid search? #hybrid" target="_blank" rel="nofollow noopener">weaviate.io/developers/wea…

@devorein @weaviate_io I am a bit confused about why OpenAI marks my embeddings with equal tokens and prompts because the company descriptions should be longer pieces of text and therefore use fewer tokens. But this happens under the hood in @weaviate_io, so maybe they can explain?

@devorein @fireship_dev @weaviate_io @ycombinator During all my testing and development, I paid max. $1, and specifically for the Y Combinator dataset, it's probably ~$0.20. Search results are okay but could be better. I think this mainly happens because not all companies have proper long-form descriptions

(🧵7/7) Using the GraphQL graph provided by Weavite, we can easily search for companies using the power of vectors. Send me YC company or search prompt and I'll show you the results/closest relatives! Or do it yourself: github.com/daandegroot123…

(🧵6/7) This industry colour-coded screenshot shows how "B2B Software/Services" (dark blue), "Healthcare" (light blue) and "Consumer" (pink) form large groups, with "Education" and "Industrials" (light green-ish) in-between. "Finance Tech" is chilling on the bottom (green).