> Technical: I started this project when the first LLMs came out. I've built extensive internal evals to understand how LLMs are performing. The hallucinations at the time were simply too frequent to passthrough this data to visitors. However, I recently re-ran my evals with Opus 4.5 and was very impressed. I am running out of scenarios that I can think/find where LLMs are bad at interpreting data.
It's nice to see an AI-centric Show HN product that uses proper evals and cares about data quality.
How did you build your initial data set that you're using for the evals? Bootstrapping a high quality data set is one of the hardest parts of really knowing how an AI product is performing.
That's the plan. I don't intend to own any stock. I want to focus on covering the broadest range of supplements across all of the marketplaces, having the richest data about them, and then focus on the affiliate revenue.
The affiliate revenue can be anywhere from 5% to 10% depending on the affiliate partner. Considering no overhead of support, inventory, or logistics, it's a pretty good deal for me, especially for now, while I'm still a solo founder.
I am more of a geek than a business shark, but when I was doing market research, I couldn't believe just how big some of the supplement marketplaces are. iHerb ($2nb+), Amazong ($8bn+), GNC ($2bn+), Vitamin Shoppe ($1bn+). However, it's a market that requires a lot of investment to enter (logistic/storage/support). I'm hoping I'm going to get my foot across the door by providing the most comprehensive solution for education, as well as discovering and comparing supplements across all of these marketplaces.
I'd love to see a project that actually analyzes every supplement on the market to make sure it actually contains what it claims to, contains it at the listed dosage, and to show anything else found (heavy metals for example). That's not something AI can do for us though since it'd involve physically collecting and testing samples.
At the moment, my focus is on what I can do by aggregating and analyzing the available data. Extracting ingredients, normalizing them, normalizing quantities across every supplement, etc. This has already proven to be a lot bigger undertaking than I could have imagined at the beginning of this journey. I had learn a lot about databases, scraping, LLMs, and evals. But it has been a tremendously fun (if sometimes overwhelming) journey.
First, I need to figure out how to monetize what I've built so far. And maybe I cannot, in which case I will start over. But I am trying to find my niche that people uniquely value. So far I found that there are is a pull from people chasing deals (e.g. finding products containing specific ingredients with the highest price per mcg) and people seeking for niche ingredients. This is not exactly the target audience I had in mind when building this, but I am glad they are finding value.
Evolving into performing actual lab tests and producing our own high-quality supplements is my dream.
Doesn’t that kind of exist already across a couple sites? Thinking consumer labs does this across lots of supplement categories. I have seen a few others too.
That is odd. When I searched for Vitamin D, it showed me a list (which I could sort by $/mg). That list included Vitamin D, Vitamin D2 and Vitamin D3. I assumed (incorrectly it seems) that this meant that it was doing some sort of prefix match. It seems that the result table highlights the "Ingredient" column, so I assumed (incorrectly) that it was searching that column. However, it doesn't seem to search the "Supplement" column, nor is it doing a prefix match (or substring match) on the Ingredient column. This is just confusing.
There is even a Google Sheet that is generated based on the data that we have, although I just noticed that it's not updating properly, so I need to fix that.
Failed to execute 'removeChild' on 'Node': The node to be removed is not a child of this node.
Something broke. We're working on it, but in the meantime, try reloading the page. If that doesn't work, come back later.
Error ID: b846e2e2ba3b483ab93f10e72ef76820
NotFoundError: Failed to execute 'removeChild' on 'Node': The node to be removed is not a child of this node.
at ds (https://pillser.com/assets/entry.client-DWgmqxdv.js:1:112074)
at gs (https://pillser.com/assets/entry.client-DWgmqxdv.js:1:113602)
at ys (https://pillser.com/assets/entry.client-DWgmqxdv.js:1:113850)
at gs (https://pillser.com/assets/entry.client-DWgmqxdv.js:1:113728)
I appreciate you letting me know. I can see it in Sentry as well. It's odd though, it looks like it's coming directly from React. There are no traces in my own codebase. I'll try to isolate it.
I don't want to, and I don't plan to. At a time when this happened, I was deep in my studies preparing for exams, and I just remember thinking to myself, "I cannot afford to pull myself into what could become a legal matter" I am now in better position to allocate time and attention should such claims be made again.
That said, the couple of brands that were removed were not brands that I would have wanted anyone to buy anyway. Not much is lost by not having them on the website from the perspective of fulfilling customers' journeys of finding a good product.
I do like the suggestion made in one of the comments informing customers why some brands are not visible on the website. That warning might on its own deter others from making such claims.
I would love to get your feedback on what data to prioritize and how to make it easier to browse.
I've learned from my last sprint that this project can get overwhelming. So now I am making it my priority to build a community of people I can reach out to for feedback and direction. If supplements is your thing, I would greatly appreciate being able to chat with you from time to time (even if it just a quick email!)
Hi, not into supplements, but really appreciating your approach to building a community and the reasoning behind it (and that you're being so transparent about it!). I'll give this a go too, thanks for the inspiration :).
Sort've related, but here in Australia pet food manufactures are not required to list the nutritional content of their foods, whereas in the US as I understand it they do.
1. dangerously unwise area to apply "AI" given all its known problems, and whats at stake
2. I tried using that site and immediately saw problems and broken behavior. The "Ask AI" feature (sounds pretty key?) literally did nothing. and doing a search for say aspirin yielded results that... had no aspirin! no brainer to bail out fast
I've fucked myself up so much over the years with supplements. Wish I could go back 30 years and tell myself to just eat real food. B6 toxicity for instance is crippling. That being said these days I just take two things: vit K2 and magnesium.
That's solid advice for most people in the Western world. K2 and magnesium are sensible choices–genuinely hard to get enough of through diet alone. The bigger problem is that most supplement sales are marketing-led. That's what pushed me to at least try to make the whole process more scientific.
It's nice to see an AI-centric Show HN product that uses proper evals and cares about data quality.
How did you build your initial data set that you're using for the evals? Bootstrapping a high quality data set is one of the hardest parts of really knowing how an AI product is performing.
The affiliate revenue can be anywhere from 5% to 10% depending on the affiliate partner. Considering no overhead of support, inventory, or logistics, it's a pretty good deal for me, especially for now, while I'm still a solo founder.
I'd love to see a project that actually analyzes every supplement on the market to make sure it actually contains what it claims to, contains it at the listed dosage, and to show anything else found (heavy metals for example). That's not something AI can do for us though since it'd involve physically collecting and testing samples.
That's where I want to take this project.
At the moment, my focus is on what I can do by aggregating and analyzing the available data. Extracting ingredients, normalizing them, normalizing quantities across every supplement, etc. This has already proven to be a lot bigger undertaking than I could have imagined at the beginning of this journey. I had learn a lot about databases, scraping, LLMs, and evals. But it has been a tremendously fun (if sometimes overwhelming) journey.
First, I need to figure out how to monetize what I've built so far. And maybe I cannot, in which case I will start over. But I am trying to find my niche that people uniquely value. So far I found that there are is a pull from people chasing deals (e.g. finding products containing specific ingredients with the highest price per mcg) and people seeking for niche ingredients. This is not exactly the target audience I had in mind when building this, but I am glad they are finding value.
Evolving into performing actual lab tests and producing our own high-quality supplements is my dream.
Costco (https://www.costco.com/p/-/kirkland-signature-extra-strength...) sells Vitamin D at less than half that price. On Amazon, the two pack of those is even cheaper.
Just an observation.
https://pillser.com/search?s=dfbtbc9110&q=%22Vitamin+D3%22
The cheapest price on Pillser for D3 is $0.26/mg
https://pillser.com/search?q=%22Vitamin+D%22
If you want to look for the specific ingredients, you can do so by clicking on the "Ingredients" of the product example: Here are the links for the Vitamin D and D3. https://pillser.com/vitamins/vitamin-d and https://pillser.com/vitamins/vitamin-d3
There is even a Google Sheet that is generated based on the data that we have, although I just noticed that it's not updating properly, so I need to fix that.
Cripes.
That said, the couple of brands that were removed were not brands that I would have wanted anyone to buy anyway. Not much is lost by not having them on the website from the perspective of fulfilling customers' journeys of finding a good product.
I do like the suggestion made in one of the comments informing customers why some brands are not visible on the website. That warning might on its own deter others from making such claims.
I've learned from my last sprint that this project can get overwhelming. So now I am making it my priority to build a community of people I can reach out to for feedback and direction. If supplements is your thing, I would greatly appreciate being able to chat with you from time to time (even if it just a quick email!)
I am at [email protected]
That means I can dump woodchips into capsules and sell them as Multivitamins with 12 vitamins & minerals, and nobody would be the wiser.
There is more rigorous testing being done in underground steroid + peptide communities than in legal nutritional supplements.
Crazy world where you can trust vialed peptides from China more than something you bought on Amazon...
2. I tried using that site and immediately saw problems and broken behavior. The "Ask AI" feature (sounds pretty key?) literally did nothing. and doing a search for say aspirin yielded results that... had no aspirin! no brainer to bail out fast
That's surprising because it seems like you just have a review site? What's the issue? Or is it just bs threats?