lilouartz

Hey everyone! Thank you for your feedback.

Whether it is positive or negative, I do appreciate it as it helps me to learn and improve the product. I really didn't expect this to get any attention, let alone dozens of comments!

To clarify: This was originally designed to help me auto migrate URL schema. I am learning as I develop this website, and SEO has been one of those vague topics where there are few hard rules. I wanted to leave space for experimentation. As I rolled it out, I became intrigued with how it functions and wanted to share my experiment with you to get feedback.

Based on the feedback, I plan to change the logic such that:

- I will track which URLs are associated with which products - If user hits 404, I will check if there was previously a product associated with that URL and redirect accordingly - If it is a new 404, I will display a 404 page which lists products with similar names

I appreciate everyone hopping in to share their perspective!

Hey everyone!

I am sure everyone is already familiar with Examine. It is perhaps the biggest resource on supplement research. The way they've been operating is that they focus on a few health areas and then scout for research papers that mention related health outcomes, and then describe the link to different supplements. This is a hugely laborious undertaking, but now with AI, many parts of this process can be automated, and that's what I am building.

Here is an example of all research that is linked to Increased Short-Chain Fatty Acid Levels:

https://pillser.com/health-outcomes/increased-short-chain-fa...

I've now indexed thousands of research papers and extracted insights from them that link those studies to different supplements available on the market. The long-term goal is to create the first of the kind supplement store where every supplement on the marketplace is linked to research. Unlike Examine, I intend to make all research summaries public and instead focus on affiliate revenue from the related supplement sales.

The biggest challenge is data accuracy. Due to these being complicated topics, I am avoiding extracting any insights beyond demonstrating a link between the study, health outcome and substance. Users are then linked to the actual research papers where they can building confidence in their decision. However, over time, as AI models advance, I am planning to expand this to a complete insight engine.

I am fascinated by the problem domain, and more generally data normalization problems and applications of LLMs to data normalization problems. I am now doing it as a hobby that I foresee to be a money losing activity for a long time into the future, but I also think there is a fair chance that Pillser could emerge as a preferred website for people who want to buy supplements due to the unique combination of science and inventory.

I am probably a few months away from having the complete database, but wanted to share for early feedback. The website is called https://pillser.com/, and you can already search for different health goals and associated research directly from the landing page.

isn't before and after a photo of different plants?

That's a very beautiful app

I am still figuring out the positioning, but the idea is to allow search among thousands of supplements and compare the matching ingredients in a spreadsheet-like interface.

In this example, we see every supplement that contains "magnesium citrate", price of the supplement, totals contents, servings, amount per serving, and price per mg.

This way is the quickest way to figure out which supplement to buy if you are looking for a specific ingredient, as it tells you the price and quantity without needing to browse every product individually.

I've also collected attributes that help to narrow down supplements based on the attributes such as gluten free, vegan, etc. and certifications such as GMP Certified, etc.

I geek out about LLMs and data normalization. I wanted to build something in an intersection of my skills and something that would appear as borderline absurd but have counterparts with billions of revenue (iHerb). I bumped into iHerb and the problem by pure coincidence as I got frustrated trying to buy supplements for myself. It started with a spreadsheet (https://docs.google.com/spreadsheets/d/1qnhEhBPkpI6rY2kdXo5z...) and quickly escalated to what you see today.

Thank you

so many people helped me a long the way. Especially Remix community. Without their Discord community, it would have taken me twice as long to build this.

Isn't it just shift+cmd+v to paste as plain text?

It is an interesting name, but I would personally not pick it for my project, even if it was in the "dev" space. It feels like branding it would be quite tough.

What's a good way to contact you?

I have a product question regarding one of the comments you've made in 2022.

I added my email to my profile if you could please email me.

I forgot one more option I considered: Even though every ingredient is stored in a different table, I could enforce unique PK across tables. Then I would not need to specify type in the ID. This would mean that I need to lookup every ID to know which tables it belongs in though.

In my case, it is the latter.

I very intentionally figured that I want to "retain" my problem so that I could test with my own product whether it at least addresses my original problem.

That being said, I am a bit nuts! I can go into decision paralyses if I don't know all the variables.

I asked myself a lot of the same questions, and also have come to similar conclusions that I just need to throw it out there to see if people are even remotely interested.

I would like to hear from others how true this is.

My understanding (based on the wording that OpenAI itself uses) is that it still uses OCR tech behind the scenes. It seems like they have developed some models to better comprehend the same OCR scanned content that others produce.

For context, I am building a supplement store.

My idea is build a supplement store that allows discovery based on science, rather than marketing. You can search supplements based on their ingredients and see them in the context of research papers about it.

I am curious how this translates to what you are suggesting.

How does this work when the product is free to use?

Yeah, I am struggling a bit with this analogy of a car.

I am building a what I can only describe as "Supplement discovery based on science and not marketing". It requires that I create a database of supplements available on the market, then allow to compare them at a ingredient level to understand which is the best one to buy.

I started building this out of frustration of trying to figure out which supplement I should buy myself. It took a ton of time to build a comparison table ( Here is one example https://docs.google.com/spreadsheets/d/1vQcLEAPk5lcg-4YR_dwD... ).

So I validated with a spreadsheet that people find it useful. And then I started building the data model for representing this across all supplements. And this is where things have gone a bit south. Somehow now I have most ingredient data at a molecular level. I tell myself that I needed to go this deep to ensure data quality (def don't want to accidentally misrepresent ingredients).

I am over this hump now, i.e. I have indexed and categorized every ingredient of over 10k+ supplements. Now I am just building the UI to compare them, but it feels like dug myself into a whole for a good part of the week. Will see what happens once I make this public; if people use it would be hugely rewarding.

But going to the car analogy... my mind is telling me that doors, a windscreen, or bike rack are all nice to have things that translate to various compromises in the UI. Just keeping that extremely simple. But that data integrity is not something I can compromise on (it is like letting car for a ride with an engine that may break).

Don't know how this way of looking at things is true for others. Sometimes I feel like I just have an OCD specific to designing the underlying data architecture.

Twice the length from the middle to the end

Ended up dumping it all into a database schema. I realized that in order to do comparison, there is quite a bit processing that needs to be done to normalize data. Things like identifying chemical compounds, nutrients, etc.

This def turned out to be a lot more cost/time consuming than I expected, but I do enjoy the learning process.

Since the data will be in PostgreSQL db, I will be able to slice and dice it however I want to export it later to Airtable or other formats.

Thanks for all the recommendations!

Thanks so much for the tip!

Do you happen to know if Airtable documents rank on Google?

One of the reasons for preferring Google Sheets is because I know from experience that I come across many public Google Sheets through Google. I don't recall coming across Airtable documents, but it might be just because there are fewer of them.

I am giving this a shot.

I have been hacking together the initial prototype. I am going to start by just aggregating the data and at first distributing it as Google Sheets document. I figured, this will be good enough experiment to understand whether people even want this as a service.

Well, not quiet. At the end of the day, the value that it would provide is a comparison. The links would be to Amazon affiliate program, i.e. I am indifferent which product you end up buying.

The end result would be a website that provides a comparison similar to how this spreadsheet does it for Omega 3 supplement https://docs.google.com/spreadsheets/d/1vQcLEAPk5lcg-4YR_dwD...

This is to say that ads and sponsored listings would play little role in something that effectively allows to compare supplements based on their ingredients.

I have no doubt that it would be useful to some people... My bigger concern is if there is a business here worth pursuing. Not exactly cheap to run (would require a lot of manual input and/or AI). Not a lot of defensibility (could be scraped by someone).

So, not quite the answer to your question, but the idea came because I saw people create spreadsheets of supplement comparisons on Reddit with each other. and I thought "wow, that's actually very useful, why isn't there an app for this"