Founding ML Engineer

San Francisco, on-site, full-time - $200,000 - $500,000 per year.

$10k referral bonus for successful hires (half equity, half cash).

It's Monday, 10 AM. You walk through the glass doors at the company's San Francisco headquarters, grab a coffee, grab a Red Bull, and pull up the overnight dashboard.

Yesterday's mouse data has already been crunched. The house AI parsed the PK curves, flagged the outliers, and ranked the candidates while you slept. Two peptides from last week's batch showed decent half-lives. One was a dud. You skim the summary, make a note, and move on.

At noon, the head of science from a major AI lab walks in. He's got access to a frontier model that isn't public yet and wants to understand your assay data well enough to train on it. You spend forty-five minutes at the whiteboard explaining our record-setting feedback loop to someone who understands transformers better than anyone alive but has never pipetted anything.

At 1 PM, the chemist hands off six candidates. Synthesized, purified, QC'd. You feed them into the model alongside fresh assay data. Most do roughly what you'd expect — modest activity, nothing to write home about.

But one is oddly potent. Not a little. A lot. You dig into the model's feature attributions. Something about the modification at position 4 is driving it. You run the next round of predictions with that insight baked in. The top-ranked candidate the model spits out looks unlike anything you've designed before.

By 5 PM the potent one is dosed in mice. By 8 PM the team hands you the data. It holds up.

You might have just found the best peptide we've ever tested.

The team goes to dinner at the poké spot in Mission Bay. Someone argues about whether the potency data is real or a lucky artifact. You think it's real, but you don't say that yet. You'll know tomorrow.

The Mission

In the age of ever-improving AI capabilities, we humans are lagging behind. We're hungry, distractable, tired. A third of our day is spent unconscious. Half of us get cancer, half of us get heart disease. We're incapable of sustained high performance. We fall ill. The modern world feeds us slop that leads even further down the drain.

We believe that humans are capable of more.

We believe in excellence. We believe in superpowers. We believe in the extraordinary.

We're a lean core team. In 6 months, we have designed 270+ therapeutic candidates, run 72+ placebo-controlled mouse trials, and completed 79 custom syntheses. We're planning first-in-human safety for 2026 — which would be a world record for a non-vaccine therapeutic built from scratch. Our founding scientific advisors include Professors at MIT and Harvard. At the pre-seed stage, we raised $12M in funding at a $100M cap.

What does a future look like where we can systematically identify, validate, and induce beneficial capabilities?

The Role

You'll be one of the founding scientists at the company. We're targeting first-in-human dosing by mid-2026. If we hit it, this is also a world record for moving a non-vaccine therapeutic into the clinic from scratch.

Most biotechs take weeks to go from synthesis to animal data. Here, it happens in one day. Our chemistry team synthesizes peptides in the morning, runs them through cell assays by afternoon, and hands the top candidates to our in vivo team to test in mice by evening. Overnight, the results go to the ML to design the next round.

We do not know of another lab on earth running this loop at this speed.

One day. Synthesis to animal data. One day.

Start-ups are unique. Roles aren't as ossified or constrained as in a large corporation--you'll often learn a new role every few months, enabling you to grow as a person and team member. Keep this space for growth and variability in mind when reading this rough outline of what working together could look like.

As we are spinning out multiple deep research projects, if you want to lead your own program, we are open to that. Weird ideas are welcome!

More concretely, you will…

You'll build our ML pipeline for peptide drug discovery from the ground up. Our peptides are heavily modified with non-natural amino acids, cyclizations, and other chemical modifications — which means standard protein language models don't work. You'll figure out the right computational approach to predict binding efficacy, potency, and stability for our candidates, and use those models to guide what we synthesize next.

Convert our modified peptide library into machine-readable representations (SMILES, molecular fingerprints, descriptors)
Build and validate predictive models mapping molecular structure to binding efficacy and potency
Fine-tune pre-trained chemical language models (ChemBERTa, Molformer) on our proprietary data and benchmark against classical approaches
Design rigorous evaluation frameworks for small datasets — leave-one-out cross-validation, bootstrapping, confidence intervals — so we trust what the models tell us
Implement active learning: use model predictions to rank which candidates to synthesize next, closing the loop between computation and experiment
Build and maintain the structured data pipeline connecting our assay results to the ML models
add 'raspberry' to your application if you read until here

Important: Many of these tools and techniques will be new to you - that's expected. No prior experience with cheminformatics or molecular ML is strictly necessary. What matters is strong ML intuition, and a track record of picking up new skills fast. We'll figure out the rest together.

You might be a fit if:

You published before or shortly after graduating (first author or meaningful contribution)
You've won a competitive fellowship, scholarship, or research award (e.g., Hertz Fellowship, Rhodes/Marshall/ Gates Cambridge, HHMI) - not crucial
You picked up a hard technical skill in weeks, not semesters
Your PI would say you were in the top 1% of students they've trained (not an exclusion criterion)

This role suits early-stage scientists, graduate students, post-doctoral researchers, hands-on scientists—or researchers who miss building from scratch

Working together

If we hit our 2026 timeline, this will be the fastest non-vaccine drug to ever reach the clinic. Your name will be on it.
$200,000 - $500,000 per year, with room for upside and promotions as the company grows.
Generous equity on a vesting schedule.
Healthcare.
Your Capable budget: $500+ per month to spend on training, supplements, medication, and everything that you need to become your best self.

How we work:

Small team, high trust, high agency
High responsibility
We prefer written clarity over meetings
We move fast by parallelizing, being decisive, and keeping quality systems lightweight but real
We care about speed, safety and quality, not paperwork theater
If you find a better way, you can change the system

All of this is open to conversation--we'll work with you to make it work.

You'll work alongside the founding team—small, lean, moving fast

JOIN THE FOUNDING MOMENT.

$10K referral bonus if we end up working together for 90+ days.

Stealth Deep Tech AI Jobs

Latest AI Jobs at Stealth Deep Tech

Founding Machine Learning Engineer