HN user

koller-m

4 karma

https://mikekoller.dev/

Posts2
Comments6
View on HN

Are you an expert in the field? Given that I have watched a lot of self-help content with a focus on dating and relationships, the app is very good at answering questions. Even extremely verbose questions.

One pain point is getting the product in the hands of people looking for answers to those questions. And wondering if the advice feels cheap because it comes from an AI and not a human.

Also, as a builder you tend to overlook simple UI stuff. The site uses Magic Link to login. You input your email address and then you're sent a link that allows you to login. Just a clarification because I've seen a lot of people input their email but not actually login today and I'm wondering if there's some confusion.

I look forward to your feedback. Thanks.

Not exactly. It uses a vector database, Pinecone in this case, to store relationship and dating data that has been scraped from the web. When you submit a question, it uses semantic search on that query to pull those most relevant piece of data, which it includes in the prompt to answer the question. Does that make sense?

Admira in six words or less: Instant advice from an expert knowledge-base

This is my second project using a vector database and OpenAI's API's. I initially created Startup AI which took the entire Y Combinator Podcast and generated answers to startup questions.

Startup AI was sort of a litmus test to see if I could build something like this. I even used Startup AI to help generate the idea for Admira. The advice being to look for an industry that is underserved by tech/startups.

I think we all run into relationship and dating problems at some point in our lives. I went through a bad breakup in 2020 and remember going through hours of podcasts and YouTube videos searching for answers on why it happened and eventually how to heal moving forward. Something like Admira would have saved me a ton of time and put me back on track a lot quicker.

I apologize for the email verification. One thing I learned from Startup AI is that bots will submit your form, which in this case costs money.

Looking for as much feedback as possible. Thanks!

I was planning on writing a more detailed blog post. I think one of the coolest things about it was the vector database and semantic search. Had no idea those things existed. Thankfully, I had done a few projects with PyTorch, so getting up to speed wasn't too bad.

Yeah, I really wanted as much feedback as possible and being new to HN, had no idea what Show HN was. I'll consider re-submitting.

Thank you for the feedback!

Inspired by Huberman AI and All-In On AI. I used Golang to download the episodes in .mp3 format. OpenAI Whisper to transcribe. And a combination of Pinecone and OpenAI API's to generate the answers. Frontend is NextJS and Tailwind CSS. As someone who is still trying to land their first full-time developer role, this project was challenging and very rewarding. I really think gone are the days when we have to digest content on 1.5x or 2x speed just to get through it all.