Building DigDep.com: A Dev’s Quest to Open Source Supplement Science

If you’ve ever searched for “best supplements for arthritis” or tried to decode ingredient lists on health blogs, you’ve probably landed on Examine.com or Healthline-style articles. They’re useful—but often limited by slow updates, paywalls, or one-size-fits-all summaries.

That’s exactly the problem I’m trying to solve with DigDep.com — a developer-led project to map supplement products directly to clinical research, using AI pipelines and transparent data logic.

🧪 From Ingredients to Research in One Click

Take this for example:

NOW Supplements, Glucosamine & Chondroitin with MSM – Joint Health & Comfort

On that page, you’ll find:

A list of relevant clinical trials on glucosamine, chondroitin, and MSM

Direct citations to PubMed and other research databases

A breakdown of which studies link the supplement to outcomes like reduced joint pain or improved mobility

User reviews, so you can contrast anecdotal experiences with peer-reviewed findings

It’s not just a product page — it’s a research navigator with structured science behind it.

🤖 The AI Behind It

I use a multi-model LLM pipeline to parse research papers, identify connections between ingredients and outcomes (like “arthritis relief”), and then validate those connections with human-like accuracy.

The Stack (Simplified):

Discovery: Lightweight open models scan abstracts for substance–outcome–dosage signals

Validation: GPT-4 or Claude reviews excerpts to eliminate false positives

Summary Matching: A final model cross-references the claim against the research excerpt

All this data is normalized across thousands of entries, so users can go from health goal → compound → product, or the other way around.

🧠 Why Not Just Use Examine?

Because Examine doesn’t link to actual products, and doesn’t let you filter for clinical evidence per product.

DigDep does.

Also:

Examine is paywalled; DigDep is free

Examine is slow to update; DigDep refreshes regularly via automation

Examine doesn’t map individual supplements to reviews and research; DigDep is built for it

And as developers, we can appreciate when a system is built modularly, using pipelines that evolve as the models get smarter.

🧱 It’s a Work in Progress, but Already Useful

So far, I’ve indexed:

20,000+ research papers

Hundreds of common health outcomes (e.g. arthritis, anxiety, weight loss, ADHD)

5,000+ supplements, matched by ingredients and dose

Each listing gets smarter as new research is added. The ultimate goal?

To make DigDep the most trusted and usable research-backed supplement directory out there.

💬 Try It and Tell Me What’s Missing

Here’s that example again:

NOW Glucosamine & Chondroitin – Arthritis Research & Reviews

If you're into LLM applications, health tech, or just curious about turning messy biomedical data into structured, navigable knowledge — I’d love feedback or ideas.

This is open-source in spirit (and maybe soon in code too). If you'd like to collaborate, critique, or just discuss model design — hit me up.

Thanks for reading 🙌