Trust Me, It Binds

Looks like science. Funds like science.

Welcome to Latent Therapeutics - a biotech that replaced their chemistry R&D with a generative model trained on textbooks, benchmarks, conference abstracts, and several thousand social-media posts. The outputs are almost perfect. Unfortunately, “almost perfect” does not beat the benchmarks. Management’s revolutionary fix: Chemist-in-the-Loop. That is you: the last human chemist in the building, hired to keep the slide deck green while investors still believe in the platform.

  

The Game

A molecular-design simulator with deckbuilding mechanics. You have a molecule, a target spectrum, and funding to justify. Play reaction cards to change both the structure and its six spectral bands; the closer you get to the target, the better the score. Superstitions turn the chemistry into multipliers. Experimental validation is too expensive. The score is enough.

The Story

Five investors stand between you and the IPO. Fortunately, they are exactly the people you would want making decisions about science.

Angelica Seed sold an iPad arcade for dogs and has been an expert in everything since. 

Monty Momentum is a biotech influencer who believes chemistry is just atoms networking. 

The Sulphurist is old money, legacy  and occult aspirations. 

The Claimant sees every competitor patent as a list of opportunities they forgot to own.

And then there is I.P.  -  He does not take meetings. The others mention those initials carefully. Survive the first four and you may get an introduction.

The Science

The joke gets better when you realise the chemistry is real.

The molecule-generation engine is powered by mlconfgen, a real machine-learning model built on the same class of algorithms used by actual AI-biotech companies. And it runs locally on your computer while you play. Cool, huh?

This, frankly, makes the whole thing much funnier. Not that the bar was particularly high.

Every chemical action in the game is legitimate, every card actually changes the molecule, and at the end of a run you can inspect everything you created. There is a very good chance (more than 99.3%)  the most of those molecules have never existed anywhere before — which is either exciting or exactly the sort of sentence that gets a startup funded.

You can read about mlconfgen, and how it works  here.

If you clicked  that link, you either have a PhD or are approximately as boring as I am. Possibly both.

Thanks to Azat Davliatshin for porting the core ONNX generative model to JavaScript, making a platform-agnostic implementation possible.

The Project

The game is currently in Alpha, and we’re very keen to hear your feedback, comments, complaints, bug reports, and any other evidence that something has gone horribly wrong.

We’re actively developing the game and aim to release updates every two weeks, with fixes, balance changes, new content, and general improvements.

Please reach out — especially if you find something broken, weird, confusing, or unexpectedly fun.

🎵 Listen to the Soundtrack 
SoundCloud