I stumbled on your Substack randomly, and I'm glad I did. I have been going through a similar ordeal -- using AI to create a case summarization system -- and I completely relate. You're much farther along (my tool is still 3 Python scripts), but now, you've inspired me to put this together on a webpage somehow.
Thanks for sharing what you've learned, especially the challenges of dependency hell.
I’m so glad you found my journey relatable and took something useful from it. Thank you! And you're absolutely right, putting your tool together on a webpage is a whole different experience.
Even when you know the exact steps, actually doing it brings a unique set of challenges and learnings. Wishing you all the best with your project!
I also remember the time I had to learned a lot about hosting, DevOps, full-stack development and running a business at the same time.
Haven’t felt so alone 😀
It gets better, though. This accelerated learning pay dividends in the future, trust me!
Maybe be too late to give advice here, but if you are in the market for cheap and easy hosting, I have had success with services like Railway and Fly.io. Hyperscalers like AWS and GCP are incredibly sophisticated and a bigger bite than you can chew if you want to roll something out fast.
Congrats on your launch! I have the opposite problem as a non-tech person. I can map out the flows (I am a UX and Product Strategy Consultant) but my MVP can only go as far as no-code (such as Claude or Bolt.new) could help me. To actually build and launch a real app means hiring - a significant investment.
Thank you for the comment, Evelyn! It’s so interesting to hear your perspective from the opposite side. Maybe we can connect more closely? Who knows, we might create something exciting together.
Thanks for sharing, Jenny! I really enjoyed reading this - this sounds like a RAG system for image searching where your knowledge base is a collection of embeddings transformed from images, is my understanding correct?
Thanks for the comment! While there are some similarities, this isn't technically a RAG system. The app uses OpenAI's CLIP (Contrastive Language-Image Pre-training) model to perform semantic image search. CLIP creates embeddings for both the text query and images, then compares them using cosine similarity to find the best matches.
Oh that's really interesting, thanks for your response! Did you have to set up a vector DB to store the embeddings then? That's the part that I'm stuck on - for context, I've been tinkering on creating a RAG Q&A app for social media posts
Actually, this implementation doesn't use a vector database! It uses OpenAI's CLIP model (via the transformers library) to compute embeddings on-the-fly for both the search query and images, then compares them directly using cosine similarity.
For your RAG Q&A app with social media posts, you'd definitely want to use a vector DB since you're dealing with a persistent collection of text. I heard that ChromaDB as a good starting point, it's pretty beginner-friendly for RAG applications.
So cool you’re pursuing and sharing these experiences. Such a wild world that is foreign and terrifying for most. You providing insight, perspective, and pathways to potential success is beyond awesome.
Thank you so much for the encouragement and thoughtful feedback! I truly appreciate it. I’m always open to hearing more comments or constructive thoughts on the product!
I stumbled on your Substack randomly, and I'm glad I did. I have been going through a similar ordeal -- using AI to create a case summarization system -- and I completely relate. You're much farther along (my tool is still 3 Python scripts), but now, you've inspired me to put this together on a webpage somehow.
Thanks for sharing what you've learned, especially the challenges of dependency hell.
I’m so glad you found my journey relatable and took something useful from it. Thank you! And you're absolutely right, putting your tool together on a webpage is a whole different experience.
Even when you know the exact steps, actually doing it brings a unique set of challenges and learnings. Wishing you all the best with your project!
Congrats on the launch! I will check it out shortly
Thank you for the visit Zan!
I also remember the time I had to learned a lot about hosting, DevOps, full-stack development and running a business at the same time.
Haven’t felt so alone 😀
It gets better, though. This accelerated learning pay dividends in the future, trust me!
Maybe be too late to give advice here, but if you are in the market for cheap and easy hosting, I have had success with services like Railway and Fly.io. Hyperscalers like AWS and GCP are incredibly sophisticated and a bigger bite than you can chew if you want to roll something out fast.
Thanks for sharing your experience here! I would love to try out the services you had success on for my next projects.
Congrats on your launch! I have the opposite problem as a non-tech person. I can map out the flows (I am a UX and Product Strategy Consultant) but my MVP can only go as far as no-code (such as Claude or Bolt.new) could help me. To actually build and launch a real app means hiring - a significant investment.
Thank you for the comment, Evelyn! It’s so interesting to hear your perspective from the opposite side. Maybe we can connect more closely? Who knows, we might create something exciting together.
I would love that! DM me and let's chat more, whenever.
Great! DM-ed you :)
Wow that's very impressive. Thanks for the pro tip, I'll try it out!
You are very welcome. I love this type of quality discussions, help us grow!
Thanks for sharing, Jenny! I really enjoyed reading this - this sounds like a RAG system for image searching where your knowledge base is a collection of embeddings transformed from images, is my understanding correct?
Thanks for the comment! While there are some similarities, this isn't technically a RAG system. The app uses OpenAI's CLIP (Contrastive Language-Image Pre-training) model to perform semantic image search. CLIP creates embeddings for both the text query and images, then compares them using cosine similarity to find the best matches.
Oh that's really interesting, thanks for your response! Did you have to set up a vector DB to store the embeddings then? That's the part that I'm stuck on - for context, I've been tinkering on creating a RAG Q&A app for social media posts
Actually, this implementation doesn't use a vector database! It uses OpenAI's CLIP model (via the transformers library) to compute embeddings on-the-fly for both the search query and images, then compares them directly using cosine similarity.
For your RAG Q&A app with social media posts, you'd definitely want to use a vector DB since you're dealing with a persistent collection of text. I heard that ChromaDB as a good starting point, it's pretty beginner-friendly for RAG applications.
So cool you’re pursuing and sharing these experiences. Such a wild world that is foreign and terrifying for most. You providing insight, perspective, and pathways to potential success is beyond awesome.
Thank you so much for the encouragement and thoughtful feedback! I truly appreciate it. I’m always open to hearing more comments or constructive thoughts on the product!