I’m building an AI tool that turns the latest PubMed studies into evidence reports. What features would you want? (Free reports on offer)

Hi all,

Some of you may remember the study summaries I’ve posted here. I used to make them by hand. I’d go through PubMed’s newest papers on a compound or topic, pick out the relevant ones, pull the most meaningful lines from each and produce a useful summary. That took a lot of time, so for the past while I’ve been building a program that does it for me. It’s now at the point where I’d like input from the people most likely to use something like it.

Here’s how it works without going too deep into the details:

  1. You enter a compound (a drug, a supplement, a plant extract or one of its constituents) and choose how many of the most recent studies you want.
  2. The program pulls those studies from PubMed, reads each one and builds a report.
  3. The report has a short executive summary, a list of unique findings written in plain language with a link to the source study for each one, and a more detailed section further down for anyone who wants the mechanisms.
  4. Findings are grouped by evidence level, so human trials, animal studies and cell work don’t get blended together.
  5. Off-label and lifespan-related findings get their own space.
  6. You can also check a compound again later and get a section showing what has been published since your last report.

Where it’s at right now

I’ve focused a lot on accuracy and on getting as much value as possible out of the studies it reads. That’s paid off, but it also means the reports can feel like information overload right now.

I didn’t want to produce a dump of mechanistic jargon. I want the report to be relevant to the reader. The mechanisms are there deeper in the report for the technically inclined, but the report opens with three sections that make it easier to take in:

  • Executive summary: a short piece of text that aims to cover everything important in the report.
  • Evidence map: an interactive visual showing where the evidence sits.
  • Unique findings: each distinct result listed once, without repeats. This gives lesser-known effects the same visibility as the well-known ones. For example, if you looked at studies on semaglutide (Ozempic), you’d find a lot about fat loss, muscle loss and diabetes. If you listed every finding from every study, you’d get 50 versions of those same results, and the more interesting, lesser-known benefits and downsides would be buried under them.

I’ve also put a lot of work into colour-coding and visuals to avoid the wall-of-text look that makes people tune out. Each section looks distinct, and the categories within each section are easy to spot.

That’s what it does today, but it’s only the first stage. I have a long list of features planned that aren’t built yet. PubMed is just the starting point, and I plan to bring in other research sources as the program grows.

I built it for people who follow this field closely but don’t have time to read every new paper. It’s meant to help you research a compound and see what the evidence actually says. It is not meant to tell anyone what to take.

Here’s a full example report on rapamycin, built from the 498 most recent PubMed records:

This is the most amount of studies I’ve run through the program for one substance!
The report is just on rapamycin, not rapamycin for longevity, or any other reason or particular condition. Some kind of functionality like this is planned in the future.

To show what else it can do, I’m happy to run some free reports. Reply with a compound you’d like covered and I’ll pick a selection to run and post the results here. Less-studied compounds are welcome too.

So I’d like to know:

  • What frustrates you most when you research a compound yourself?
  • What would you want a report like this to show that summaries and reviews usually leave out?
  • What would make you trust or distrust a tool like this?
  • Which other sources would be most useful to add, such as clinical trial registries, full-text papers or non-English research?
  • What else would you want it to do, such as comparing several compounds for the same goal?

ANY ideas are welcome, big or small. I’ll read all of them, and the best ones will go on the roadmap.

As i have paid subscriptions to various LLMs i use them to source research material. Otherwise i am interested in “news” in the sense of things that change.