yuvalkarmi

Hey folks, I'm Yuval. I run a tiny startup called Glitter AI. It's just me full-time here, with a couple of freelances to help here and there.

A couple of months ago, I went from managing zero requests to hundreds -- overnight (won Product of the day on Product Hunt). As someone who gets VERY easily distracted (maybe you relate), I had to find some sort of way of handling all the chaos if I didn't want to burn out. I came up with a pretty cool automation flow that I thought folks on HN here may be interested in reading about :)

So here goes:

Most of my interactions come through Intercom. After the product hunt launch, I got waves of bug reports and feature requests. So I link every incoming message to a webhook. I do manually trigger this step through Intercom after replying to the customer (so there's context for the AI - more on that in a bit), but then everything from there on out is automatic.

Next, Make.com grabs the conversation details and runs them through prompts that guide OpenAI to "act like a savvy product manager." The task is to act like a customer-support-rep-meets-PM: parse and summarize conversations into actionable points—be it a bug report, feature request, or feedback (I'll include a link to a post with the full prompt below).

But summarizing wasn’t enough. I needed this flow to live somewhere, and to categorize everything.

I chose Notion, but it could have just as easily been Airtable or the likes (or Excel, honestly). I call the space in Notion my "Second Brain."

Once in Notion, every request is linked back to its original Intercom conversation, which helps me quickly see patterns like repeated feature requests or pesky recurring bugs. It’s like having a high-level map of what users need.

Then in Notion I do manually quickly go through the summary and tags to ensure everything is in place. I then attach those clean summaries to existing pages using a variation of the RICE framework to prioritize my action items.

Every page counts the number of requests linking to it, so I can keep track of recurring bug reports / feature requests / product feedback etc and prioritize accordingly.

I can't include screenshots here, so I'm linking to my original post. It's mostly the same content as here, but with the prompt and a bit more detail, if you're interested: https://www.glitter.io/blog/how-i-use-ai-and-automation-to-r...

Hey HN friends! Yuval here, I'm the maker of Conversational Demos: https://www.producthunt.com/posts/conversational-demos?utm_source=hn

This is a product with a bit of an interesting history: it evolved from my fascination with what we can do with LLMs and video.

My original (engineering) thinking was "how cool would it be if we let someone search through video with natural language"?

I built the original proof of concept for this over the weekend, taped together with passion, code glue, and sheer will to look away from the really shitty code I originally wrote :)

When I showed it to users, I discovered that nobody was really willing to pay for video search - everyone thought it was cool, but there was no hook from a commercial standpoint. And so, as these things happen, I ended up turning it a marketing / sales tool. That's all I'll say about the commercial aspect of it, though I'm happy to expand on it if interesting to the community here.

Technical stuff: ---

Behind the scenes, when you upload a video (or record one - it's integrated with the not-yet-officially-released Loom SDK), it sends the video to AWS Transcribe, which works with nearly 40 languages, and at the same time converts it to an MP4 format that's tiny and viewable across devices (I do this with AWS Media Transcoder).

Next up, I break the transcription down into chunks of around 30 seconds and then send it to OpenAI's embedding database.

Through TONS of trial and error, I learned that to get good latency on this thing, you can't really use a (cheap) vector database, and so I'm doing something pretty hacky, and storing the index on a always-on Docker (deployed to fly.io) as opposed to a serverless function, which I was really trying to use until I realized it's not fit-for-purpose.

Next up, I upload the embedding index to S3 for later use, as well as store it locally on the machine. That's because the machine has ephemeral memory, and when I re-deploy, the index goes away, and so I just re-download it from S3 if it's not there.

I have no idea if this project will commercially succeed. It's live on Product Hunt now, and we'll see how people respond.

To be honest, I find building the most fun, and marketing the least fun. So on that front, I did learn several important things about LLMs and building chat-bots along the way:

1. Use the simplest solution if you're optimizing for latency

2. You can't just feed a VTT transcript into an LLM and expect it to figure out the context - you have to do your own chunking

3. There are lots of open source projects out there for using LLMs (langchain, llamaindex, auto-gpt, etc), but just using OpenAI's APIs directly yielded the best results at lowest latency, and least complexity.

I'm curious to get the feedback of the HN community - less so commercially, and moreso on the technical front, UX, etc.

What suggestions do you have, and what would love to see?

Thanks! Yuval

Hey HN friends! Yuval here, I'm the maker of Conversational Demos:

https://www.producthunt.com/posts/conversational-demos?utm_source=hn

This is a product with a bit of an interesting history: it evolved from my fascination with what we can do with LLMs and video.

My original (engineering) thinking was "how cool would it be if we let someone search through video with natural language"?

I built the original proof of concept for this over the weekend, taped together with passion, code glue, and sheer will to look away from the really shitty code I originally wrote :)

When I showed it to users, I discovered that nobody was really willing to pay for video search - everyone thought it was cool, but there was no hook from a commercial standpoint. And so, as these things happen, I ended up turning it a marketing / sales tool. That's all I'll say about the commercial aspect of it, though I'm happy to expand on it if interesting to the community here.

Technical stuff: ---

Behind the scenes, when you upload a video (or record one - it's integrated with the not-yet-officially-released Loom SDK), it sends the video to AWS Transcribe, which works with nearly 40 languages, and at the same time converts it to an MP4 format that's tiny and viewable across devices (I do this with AWS Media Transcoder).

Next up, I break the transcription down into chunks of around 30 seconds and then send it to OpenAI's embedding database.

Through TONS of trial and error, I learned that to get good latency on this thing, you can't really use a (cheap) vector database, and so I'm doing something pretty hacky, and storing the index on a always-on Docker (deployed to fly.io) as opposed to a serverless function, which I was really trying to use until I realized it's not fit-for-purpose.

Next up, I upload the embedding index to S3 for later use, as well as store it locally on the machine. That's because the machine has ephemeral memory, and when I re-deploy, the index goes away, and so I just re-download it from S3 if it's not there.

I have no idea if this project will commercially succeed. It's live on Product Hunt now, and we'll see how people respond.

To be honest, I find building the most fun, and marketing the least fun. So on that front, I did learn several important things about LLMs and building chat-bots along the way:

1. Use the simplest solution if you're optimizing for latency

2. You can't just feed a VTT transcript into an LLM and expect it to figure out the context - you have to do your own chunking

3. There are lots of open source projects out there for using LLMs (langchain, llamaindex, auto-gpt, etc), but just using OpenAI's APIs directly yielded the best results at lowest latency, and least complexity.

I'm curious to get the feedback of the HN community - less so commercially, and moreso on the technical front, UX, etc.

What suggestions do you have, and what would love to see?

Thanks! Yuval

Hey folks, question -- my buddy and I are working on a new OSS project that's been taking off recently and we're starting to gain some traction.

It's taking up A LOT of our time.

This is cool enough where we're actually considering quitting our day jobs to maintain it, but we still need to somehow feed ourselves...

I'm curious if there are other OSS maintainers here who have done it before and can give us their two cents (no pun intended) on monetization?

Do we ask for donations? patreon? just stick to sponsorships? Or is there another way we're missing (short of opening a company around it, which we'd rather not do for reasons out of the scope of this post).

Any help appreciated!

Hey HN community, this has been on my mind given the rise of AI for content generation, and the whole Elon-Twitter-bot situation, and I wanted to solicit the opinions of smart people:

To avoid having a toxic / bot / spammy online environment in different places, why don't we just verify each internet user's real identity in either a centralized / decentralized way?

It's not difficult to do identity verification nowadays - if you try to sign up for one of the scooter apps for instance, they'll verify your government ID. Some other services will use your face to know it's actually you, etc.

If we only allow signups to certain services (Twitter is one example) if someone has verified their real identity, doesn't that by default get rid of most bots? (barring criminal identity theft).

Would love to get your take!