See how much of this job your resume covers, and what’s missing.
Want a recruiter to go through it line by line?
Get professional reviewOne click makes a copy of your resume for this role, scores it against the posting, and proposes edits line by line. Your default resume is never changed.
Upload your resume and we draft a letter for this exact role, tailored to what it asks for.
You will act as the scientific authority to build and validate benchmarks for AI agents by designing complex cell therapy experiments and datasets. You will interrogate AI reasoning, adjudicate scientific conflicts, and ensure model conclusions are grounded in rigorous biological evidence.
Latch builds rigorous scientific benchmarks (benchmarks.bio) for AI agents across biology. We work with frontier labs and pharma to measure whether agents can handle real biological workflows: the messy, multi-layer reasoning that matters in the field.
We're seeking an AI Biologist to join our Cell Therapy Discovery team, pushing what AI can achieve in engineered cell design.
You will be the scientific authority on ground truth, testing whether agents can reason like an expert cell therapy scientist. The data misleads constantly: killing is not persistence, a marker is not a phenotype, and an apparent potency difference is often just a difference in how much receptor is on the cell. Knowing what the evidence actually supports is the work. You will build the tasks and datasets behind our benchmarks across CAR, TCR, and other engineered cell formats, from antigen selection to candidate nomination.
Build the task
Write gold-standard answers with the rationale an experienced cell therapy scientist would defend.
Design scenarios across the discovery arc: antigen prioritization, receptor and construct design, phenotype, potency and killing, cytokine profiling, exhaustion and persistence, on-target off-tumor triage.
Curate datasets from real programs: multiparameter flow and spectral panels, luciferase, impedance, and imaging cytotoxicity across E:T ratios, cytokine multiplex, repeat-stimulation and serial-killing assays, inhibitory receptor panels, memory phenotype distributions, normal-tissue antigen expression.
Build analysis pipelines in Python where useful, with tolerances that hold across donors, runs, and readout platforms.
Set the hard calls
Catch the artifact. A construct that looks weak because its surface expression is 20 percent rather than 90 percent. An E:T at saturation that erases a real potency difference visible only at lower ratios. A viability readout depressed because dead effectors overlay the targets at high E:T. Activation-induced cell death confounding a long co-culture. Lysis not normalized to the matched untransduced control at the same E:T. Donor variation read as a construct effect. Spillover error in a dim inhibitory-receptor channel.
Separate activation from exhaustion. PD-1, CD69, and CD25 rise transiently on recently activated effectors and prove nothing alone. Stable exhaustion needs sustained or repeated stimulation, multiple co-expressed inhibitory receptors, and functional loss. Test whether agents know which timepoint and evidence the claim requires.
Recognize tonic signaling. Antigen-independent activation from scFv instability and self-aggregation, or from hinge, transmembrane, and costimulatory domain choice, drives early differentiation and exhaustion with no antigen present. Higher affinity can raise antigen-specific and tonic signaling together. Test whether agents reach for the construct explanation instead of calling it biology.
Adjudicate conflicts. Strong short-term killing against poor persistence. Low cytokine output as a design choice or a failing product. Then name the experiment that settles it.
Weigh the tradeoffs. Potency versus exhaustion, cytokine output versus toxicity risk, affinity versus on-target off-tumor selectivity, and when an expansion gain costs a memory phenotype.
Judge the agent
Interrogate reasoning step by step, and challenge conclusions that would not survive a project review.
Pinpoint where agents go wrong: why the step fails, what an expert would do, what evidence would change the call.
Score confidence, rewarding explicit uncertainty over overconfident calls on ambiguous data.
PhD in immunology, cell biology, bioengineering, or a related field. An MS with equivalent hands-on program experience also qualifies.
2 to 6 years hands-on in cell therapy at a biotech or pharma, with your own analysis work in the last two years.
You have personally generated or analyzed data that decided which target or construct advanced, and you can name the threshold and the call.
Direct experience with at least one of: CAR or TCR construct design, multiparameter flow panel design and analysis, cytotoxicity and co-culture assays, cytokine profiling, or exhaustion and persistence characterization.
Deep in one of those, fluent across the others.
Strong experimental design, quantitative analysis, and statistics.
AI-ready from day one: daily use of Claude, Cursor, or similar, with the judgment to spot where model reasoning is technically correct but scientifically unsound, and the confidence to overrule it.
Bonus: Python or R for your own analyses. Allogeneic or gene-edited cell therapy, NK or Treg formats, or in vivo CAR.
$140K–$180K OTE (on-target earnings), with no ceiling.
Your pay is your OTE times your weekly quota attainment. Hit quota, earn your OTE; double it, earn double. Example: $150K OTE at double quota = $300K.
Your first 4 weeks (ramp) are paid at full OTE while you learn our workflows.
You’re paid full OTE whenever quota can’t fairly apply: when a new project is starting, the platform goes down, or we pull you onto a special project.
In the US: full-time W-2
Guaranteed floor (“base salary”) set by your state, paid on the 15th and last day of every month
Health, dental, vision: 100% covered for you, 50% for dependents
401(k) (no match)
PTO: 15 days to start, more as you’re promoted
Outside the US: contractor
Paid on performance. We want our team to have stability, so we pay 60% of your monthly OTE up front as an advance, balanced against your earnings in later paychecks.
US benefits don’t apply.
Two tracks, based on output.
Individual contributor: six levels, so up to five promotions, with OTE rising at each one.
Management: after your first promotion, you can move into managing a pod or keep growing on the IC track.
Location: remote or onsite at our Mission Bay office in San Francisco. Onsite is preferred: working side by side is how we share knowledge and keep our work rigorous. We have a global team, and remote teammates regularly visit the office.
Work authorization: we sponsor H-1B (new and transfers), O-1, and TN, and support STEM OPT.
We practice radical transparency: you’ll know how the company is doing, what’s working, and what isn’t. It’s always us against the problem, never each other. We’re opinionated but respectful of each other’s time, and we think through possible solutions before bringing a problem to anyone.
We use AI constantly, so we reward original expert judgment, not AI output passed off as your own. You’ll start on the bench that fits your expertise and can move across teams as you build a track record.
Outside of work: company outings most months, team off-sites, and a community that does cycling, soccer, figure skating, boxing, running, martial arts, music, books, and game nights.
Apply. If your profile matches, we’ll send you a take-home project.
Take-home project (HackerRank). We review it as soon as it arrives.
Technical interview with a Member of Technical Staff (pod lead).
Culture interview with our C-suite.
Offer.
Background check, once you accept.
Stop the endless job search. Our AI finds and applies to the best jobs for you.
Featuring 213,641+ Jobs in Others
Answer easy questions
213,641+ jobs across 15+ categories
Get your best job matches
Only hand-screened, legit jobs
Find a remote job faster
No ads, scams, or junk
“I was the first applicant for a remote marketing position that got listed on the company website the same day I applied. Had an interview within 48 hours!”