You can check it out at https://podverse.ai.

See it in action on an example podcast here: https://www.podverse.ai/podcast/omnibus

I’m a former engineer at Google and Apple, former CS prof at Harvard, and was a co-founder at Fixie.ai. I truly feel that LLMs are a revolution in our field, and I wanted to bring something to life that bridged between two deep interests of mine — AI and podcasts.

I’ve been listening a ton of podcasts lately, but it's frustrating to find relevant episodes, or even content that I had listened to previously. I also know that podcasters want to take advantage of the latest AI tech to connect with and expand their audience, but they're not usually equipped to bring all the pieces together. I wanted to build a complete, end-to-end solution for podcasters that is easy to use.

All you have to do is provide your RSS feed URL, and Podverse does the rest — automatically ingesting new episodes as you release them, generating transcripts and summaries, identifying speakers, and populating a custom chatbot that knows about all of the content of your podcast.

I built the site from scratch over the last few months (in my copious spare time!). It’s built entirely using Next.js deployed on Vercel — no backend servers beyond Vercel’s server components and API handlers. For running background tasks, like podcast ingestion and transcription, I used Inngest, which fills a big gap in the Next.js stack. Deepgram is used for ASR, which is fast, affordable, and very high quality. Supabase handles the storage, including the vector database (using pgvector). Auth is done with Clerk, which was super easy to integrate. Happy to answer any questions about the stack.

I’d love to get your feedback, and especially hear from podcasters to learn more about what features you’d like to see. Thanks!

Comments URL: https://news.ycombinator.com/item?id=41697569

Points: 4

# Comments: 1

The blog posts in question have nothing to do with Restack's own products. They seem to have generated thousands of blog posts on all kinds of technical topics. For example, searching for "Streamlit icons", I found this page:

https://www.restack.io/docs/streamlit-knowledge-streamlit-icon-list-guide

which confidently tells one to use the "st.icon()" function to add icons to your Streamlit app. Streamlit has no such API, and never has; this is completely hallucinated information that was clearly generated by an LLM.

Last week I sunk about an hour into getting a "solution" proposed by a Restack blog post working, only to discover that the approach it was describing simply does not work.

Here's a page that tells you how to deploy Streamlit on AWS Amplify:

https://www.restack.io/docs/streamlit-knowledge-streamlit-aws-amplify-integration

This page is totally bogus. Streamlit is a Python package, and Amplify requires a Node.js program. What it is describing is impossible. But the article is very confident in its tone and wording, making it look like an easy matter of running a few commands. Someone who did not understand this fundamental limitation of Amplify and Streamlit working together could sink a _lot_ of time into trying to debug why it was not working.

Shockingly, they seem to have thousands of these AI-generated blog posts on their site. Here is just the list of the first 100 (out of more than 500) articles about Streamlit that they have generated:

https://www.restack.io/docs/streamlit/kb-index/0

I find this appalling in large part because Restack is an AI company. They are clearly doing this to boost their SEO, but in so doing they are wasting the time of good people who have jobs to do and are used to relying on Google searches to surface useful, relevant content. Even worse, I fear this Restack garbage will end up in the next web crawl used to train future AI models, which will now, themselves, hallucinate back the garbage that Restack has fed them.

Restack, you can do better. Please take down your broken and misleading AI-generated blog posts.

Ironically, once Google catches on that the entire Restack.io site is spam, they will probably stop showing up in any Google searches. I hope the SEO was worth it while it's still live.

Comments URL: https://news.ycombinator.com/item?id=41587090

Points: 4

# Comments: 1

(Edit: For clarification, the NPM stats refer to the number of times the library was fetched from NPM's servers, not the number of times a developer incorporated it into a project. For popular NPM libraries, this number can be in the many millions of downloads per week. My confusion stems from the fact a library used by thousands of projects will have substantially higher numbers in NPM than what is seen here. It is of course possible that this number counts developers incorporating these icons not through the NPM library.)

I'm a co-founder at Fixie, a Seattle-based AI startup. We're launching an experiment to showcase what's possible with Fixie's real-time LLM-powered voice platform. You can check it out at https://hisanta.ai - you can have live voice chats with Santa, Mrs. Claus, Rudolph, and other friends (toggle the naughty/nice switch at the top for some other characters!)

We'd love to get your help testing this and get any feedback -- especially from people with kids :-)

All the code for this app is open source: https://github.com/fixie-ai/hisanta.ai

This is not just a thin wrapper around LLMs. To get the performance to this level, we've had to do a lot of engineering to wire together ASR, LLMs, and TTS models efficiently, and are doing some clever things to hide latency. Happy to answer any questions about the technology.

The Fixie platform (https://fixie.ai) powers the app and embeds the logic for managing voice sessions with LLM agents, which can be built either through a no-code UI or using our AI.JSX (https://ai-jsx.com) framework for building LLM apps. We support a range of ASR and TTS models under the hood as well.

We'd love your feedback! Drop me a line or join our Discord: https://discord.com/invite/tRMV7PW8f2

Thanks! Matt on behalf of the Fixie team

Comments URL: https://news.ycombinator.com/item?id=38618118

Points: 8

# Comments: 4

The main point of my article was the surprise around the extent of the VC network and the helpful interactions with them. Before going through this process, VC was a black box to me. Now, a lot less so.

A few folks below have pointed out that I must have had an extensive VC network to draw on. Not quite. I knew 3-4 VCs casually from having worked at a couple of other startups. None of them invested in us by the way. It certainly didn't hurt to get their advice. Now, of course, I have a rolodex full of dozens I could potentially call up at some point, which is useful. I can see how founders who have done it before likely find it a lot easier to get the ball (and the checks) rolling.

OctoML is changing how developers optimize and deploy machine learning models. We are the creators of Apache TVM, a popular open-source compiler that transforms ML models into highly-efficient binary code optimized for the specific hardware and model architecture. We are also building the Octomizer, a cloud service that enables developers to optimize and package their ML models through a modern web app as well as a rich API surface.

Series C, over $130M of funding, company is 100 people and growing to 200-250 next year.

ML Systems Engineer - https://jobs.ashbyhq.com/OctoML/c4728daa-c818-44e2-b56f-95af...

ML Training Systems Engineer - https://jobs.ashbyhq.com/OctoML/3fa67134-96f3-4bbe-9bc6-9e30...

Cloud Backend Engineer - https://jobs.ashbyhq.com/OctoML/4a84f36b-6c19-423a-98bf-e509...

Apache TVM Open Source Engineer - https://jobs.ashbyhq.com/OctoML/3137c2d9-af6d-4948-8111-362c...

Embedded Software Engineer - https://jobs.ashbyhq.com/OctoML/c263e74a-d74c-4940-8bbf-a65c...

Infrastructure Engineer - https://jobs.ashbyhq.com/OctoML/aa619a7e-fe22-427a-9b0f-67f0...

Associate Product Manager - https://jobs.ashbyhq.com/OctoML/02930712-15b5-42b4-b90b-539a...

Field Engineer - https://jobs.ashbyhq.com/OctoML/16ab7270-913b-4af2-bedc-3683...

If it weren't for the fact that Google makes level information public within the company, I doubt there would be quite so much emphasis on getting a promotion. Most people want to get L6 for the recognition and the fact that it elevates you above other engineers in terms of status and influence. When I worked at Apple, levels were private, so you had to treat everyone the same -- a person's influence was not dictated by what level they had on their company profile (there is no company profile at Apple). In my opinion this approach was much better, and significantly lessened the internal competition for promotions.

https://www.kleinbottle.com/

Comments URL: https://news.ycombinator.com/item?id=26993995

Points: 20

# Comments: 1

The article misses a really important point that you can't just revamp the UX for a product used by billions of users without some pretty serious blowback. Startups like Notion have a great deal of flexibility to tweak and innovate on their UX as much as they like, but even changing something minor in Gmail or Google Docs impacts orders of magnitude more people, using the product in such a huge number of environments -- phones, tablets, PCs -- in every language and every corner of the world. Every time Google has tried to make a major UX change -- look at Inbox, for example -- the challenges of bringing all of the existing users over to the new experience are very real. As a result, the UX tends to evolve in smaller steps, which (of course) results in the final result being more of a hodgepodge than you would get if you just started from scratch.

Google has very good UX designers, UX researchers, product managers, and engineers. These people know how to design good user experiences and care very much about the end result. But there is the reality of being boxed into design decisions that are difficult to undo without making some really major changes that are highly disruptive. Now, you could just say that Google should bite the bullet and hit reboot on some of its bad UX decisions from years ago. That is always an option, but it is often difficult to justify the benefits of an improved UX versus the productivity hit to all of the existing users.

https://octoml.ai/#op-398625-senior-platform-engineer

OctoML is developing technology to compile and optimize Deep Learning models for deployment on a wide range of hardware targets. We're the creators of TVM.ai, an Apache Incubator project that automatically generates highly-optimized code for an ML model.

We're looking for a senior software engineer to join our Platform team, developing cloud services for compiling, tuning, benchmarking, and packaging ML models. We program in Rust, Python, and C++.