cosbgn

I'm working on a tool to do evals for voice agents. The way it works is that you simply post your recordings, tools, instructions, etc and we do a diff to spot if something changes, then we mark a note on the change and run evals.

The main idea is that you don't need to configure anything, simply send us the data and we should figure out evals for you.

If anyone is building with realtime voice send me an email at username at Gmail and I'll try to help you improve your tool for free (In exchange I get to talk to real users)

Hi everyone, this is a very early version of what I am building: an AI accountability coach.

I know many of you will be skeptic, but in reality one of the hardest things to do is to remember to keep track of progress. A simple automation which calls me after my gym sessions and logs everything in a virtual diary could be a great help.

It's built using elevenlabs, twilio, openai (for the extraction logic) and gemini (for the call).

It's still a beta as I have some fixing to do, but it's free for the first 30 days, so feel free to try it and share your feedback with me!

I think the best would be to download also the entire wikipedia stored as embeddings. Seems like the best of both worlds.

Location: Finland

Remote: Yes

Willing to relocate: Yes

Technologies: vue.js/nuxt/node

email: cosbgn@gmail.com

I'm an entrepreneur and since I'm not breakeven yet I'm looking for part-time work. Rather than coding I would like to help with marketing. I'm fairly good with Google ads, landing pages, seo etc.

Currently I'm working on https://unfetch.com and https://rispose.com where I do everything from code to marketing to customer support.

The biggest issue I have with MCP is that some companies like notion expect every single user to host and run their own MCP. It would be so much easier for everyone if they would simply host it themselves and give access on MCP.notion.so like stripe and others do.

In the hopes that it's worthy of the HN crowd, this demo is fully open and requires no login. Let me know your thoughts and feedback!

I'm building https://rispose.com which is a tool to embed chatgpt on any site. It's super cheap and it's growing nicely.

Rispose is the first tool I build which isn't tightly coupled on another provider.

I can easily swap OpenAI for another provider or even host my AI. This IMO is something really interesting and something before was relay hard to achieve (most tools are built on top of bigger platforms).

Both, you can choose between documents (added in context) and files (RAG). It's explained fairly well in the dashboard once you login.

Airbnb is, in my opinion, one of the companies with the biggest negative societal impact.

Airbnb complementary destroyed some great places and rural cities around the world, because of them:

- Food/Restaurant prices go up

- Food/Restaurant quality goes down

- Real estate price goes up

- Dirtiness goes up

- Bugs and other animals thrive

- Quality of shops goes down, everything is disposable

- Quality of building/materials goes down, why install a wooden window if it's for an airbnb?

You could argue that it's not because of Airbnb but because of tourism, but I don't think so, Hotels are heavily regulated, they have strict cleaning rules, max occupancy and are built to host a certain amount of people, mass tourism wouldn't be possible without Airbnb.

Italy/Finland! - Fixed, thanks!

Exactly, offer something truly polished together with a very good price and ideally vitality built in.

thanks, fix is already live!

Yes, all requests pass by our server.

I run a small but fairly successful "embed chatGPT on your site" widget https://rispose.com

I'm acquiring customers by:

- Offer a 100% free unlimited solution (with branding) I get a lot of daily clicks from people coming from my customer's website

- Offer a really good price. My competitors are about 5X more expensive. I'll eventually maybe raise my price, but for now I have a lot of people switching to my tool

- Affiliates. This is something new I'm still testing.

In summary a good free product which links back to you get's you millions of requests per month!

Try https://unfetch.com (I've built it). It can handle both inbound and outbound emails

We handle over 1M requests per month using the Assistant API on https://rispose.com which apparently will get depreciated mid 2026. Should we move to the new API?

Hello everyone, I've quickly built a proof of concept here. I'm interested on tracking how different AI models perceive brands. This can be used for marketing research, SEO and more.

Here you have some public dashboards:

- Sendgrid: https://oparse.com/brands/b5uh7u1zpvzh/prompts - Coca Cola: https://oparse.com/brands/kcvcgdw191h9/prompts

If you are interested on something like this, or working with me just reach out at username at gmail.com

We are building an AI rag system and selling it to shipyards. Each ship requires thousands of unique documents to be "read" daily by big teams.

To be fair it's right criticism. I should have used indeed "m" as that's the correct unit and Mt is incorrect and confusing!

I can definitely be wrong but I wouldn't trust this map. I can't find any reference of attacks near Ibiza online.

Article is wrong. They mainly (like this one) happen in the Strait of Gibraltar on the ocean side. I think it never happened inside the Med so far.

Meters

Orca attacks don't happen in the Med. They happen in the Atlantic ocean.

I'm a skipper and I recently spoke with other skippers who do these routes often.

All orca attacks happen of 20mt+ deep waters. It's common knowledge that in the hot orca areas you must be in shallow waters (20mt or less).

As in all orca attacks so far, also this one was on a hot area and on deep waters.

I live here!

Same, I'll probably move to https://github.com/unjs/h3 since it's used anyway in Nuxt (which I use for other projects)

Hey everyone I hope you like it. I'm trying to integrate AI and APIs. It's not super simple as you have a lot of issues like:

- APIs change - Models can't access real time data quickly (e.g. scrape docs on real time) - Sometimes libraries are required - Models MUST make the right call - ETC

I'm solving this by saving API specs as vectors and injecting them in the prompts + training the AI to make API calls, + having my own code interpreter where I can run any library from NPM in real time (no install step required thanks to deno)

Let me know your feedback and ideas around this!

It's not opensouce, even though it's something I would like to be able to do soon. We offer enterprise plans, which use LLAMA2 instead of openai and can run on your own cloud (or even locally if you have enough RAM). If you are interested send me an email at c@unfetch.com

This is interesting, I do something similar with unfetch.com - I have some examples on unfetch.com/directory - There are a lot of potential use cases for LLm & APIs

Google is just so easy for login. No need to deal with password forgot, reset, email verification etc. But I'll add login via magic link soon.