Astrid is a game changing AI personal shopper and stylist. She's like ChatGPT on steroids for apparel. She can help you decide your style vibe and search tens of thousands of products so you don't have to.
Hi Product Hunt! CEO of Astrid here. I'm super excited to show everyone this product and get your feedback. As far as I'm aware, this is one of the first (if not the first) apparel personal shopping agents for consumers that is live. This is a truly agentic, ChatGPT-like experience.
We built this because shopping for apparel online has gotten pretty terrible. The variety in apparel has never been better, but unfortunately there's tens of millions of e-commerce stores. Shopping is like one long endless scroll. E-commerce personalization has never really delivered on its promise. Paid ad placements and paid partnerships on social make recommendations feel scammy.
I don't think building a product like this that understands shopping and style was possible until very recently. The access we have to state of the art VLMs, LLMs, embedding models, etc. has unlocked a world of possibilities. It's truly never been a better time to build.
Please let me know what you think! Also happy to answer questions.
We are working on way to make it faster. She does a lot of work behind the scenes which is the cause of the latency. But we have some ideas on how to fix this.
For now Astrid is only for women. We can open it up to men if there is enough interest! To make it easier to launch we chose one gender. The reason is because brands are different between women and men. There's a lot of brands that focus on just one gender. We technically have a lot of men's products in the catalog currently, but the brands are overwhelmingly focus on women. And so far getting brand data in has been one of the hardest challenges.
Congrats on the launch everyone! Curious, what are the most unobvious challenges with surfacing styles based on user queries? Is it hard to get models to have good taste when it comes to picking products?
@asim_shrestha1 Good question! The models are surprisingly competent with fashion and style. We did a major project where we evaluated nearly all frontier LLMs on this (Gemini, OpenAI, Anthropic, Llama, Deepseek, etc.). Some models are head and shoulders above the rest.
I would say what's been the most unexpectedly hard thing is finding products via image embeddings. We tried CLIP (by OpenAI) and even fine-tuned it, but still it's quite hard to get results exact enough for fashion. We even experiment with Fashion-CLIP, which is CLIP fine-tuned on Farfetch product data.
All that said, after hitting those roadblocks and having all those learnings, I've never been more confident about deep learning models helping users in the fashion and style space. I think there's a lot of possibilities and we're not far off from some major unlocks.
Congrats on this launch, this is fun to use! Really good at reiterating and adding more context behind my prompts. I'm still hopeful to see conversational AI keep up with users' more nuanced style preferences in a way other than verbal explanation. Great job Astrid team!
Astrid: Personal Shopping Agent
Really like the concept and the demo looks great, but it seems a bit slow, as you said. One question, is Astrid also for men?
Astrid: Personal Shopping Agent
@manuel_lemholt_berger Thank you!
We are working on way to make it faster. She does a lot of work behind the scenes which is the cause of the latency. But we have some ideas on how to fix this.
For now Astrid is only for women. We can open it up to men if there is enough interest! To make it easier to launch we chose one gender. The reason is because brands are different between women and men. There's a lot of brands that focus on just one gender. We technically have a lot of men's products in the catalog currently, but the brands are overwhelmingly focus on women. And so far getting brand data in has been one of the hardest challenges.
@kyle_rush Get it! Let me know when you launch for men.
Reworkd
Congrats on the launch everyone! Curious, what are the most unobvious challenges with surfacing styles based on user queries? Is it hard to get models to have good taste when it comes to picking products?
Astrid: Personal Shopping Agent
@asim_shrestha1 Good question! The models are surprisingly competent with fashion and style. We did a major project where we evaluated nearly all frontier LLMs on this (Gemini, OpenAI, Anthropic, Llama, Deepseek, etc.). Some models are head and shoulders above the rest.
I would say what's been the most unexpectedly hard thing is finding products via image embeddings. We tried CLIP (by OpenAI) and even fine-tuned it, but still it's quite hard to get results exact enough for fashion. We even experiment with Fashion-CLIP, which is CLIP fine-tuned on Farfetch product data.
All that said, after hitting those roadblocks and having all those learnings, I've never been more confident about deep learning models helping users in the fashion and style space. I think there's a lot of possibilities and we're not far off from some major unlocks.
Reworkd
@kyle_rush Amazing- can definitely see that work through the products Astrid surfaces!
Are you planning to release any benchmarks for what models did "best"/show what your grading criteria was? Would be super interested to learn more :)
Astrid: Personal Shopping Agent
@asim_shrestha1 Possibly! It was a really interesting learning exercise.