AI Biologist - Microscopy

LatchBio•San Francisco, CA

About The Position

Latch builds scientific benchmarks for AI agents across biology, working with frontier labs and pharmaceutical companies to evaluate agents on real biological workflows. The AI Biologist will serve as the scientific authority on ground truth for the Microscopy team, assessing whether AI agents can reason like expert microscopists and distinguish biological evidence from artifacts and uncertainty. The role involves collaborating with software engineers and scientists to develop tasks and datasets across cell biology, tissue, and disease imaging. The AI Biologist defines expert reasoning for imaging questions, writes gold-standard answers and rationales, designs realistic scenarios across the imaging workflow, and curates datasets across multiple microscopy and pathology modalities. The role also builds Python analysis pipelines, including deep-learning segmentation, and helps translate scientific judgment into software workflows. The position tests and adjudicates difficult or ambiguous cases, evaluates whether agents choose appropriate imaging methods, and probes their reasoning and confidence. Latch practices radical transparency and a collaborative, problem-focused culture. Team members have autonomy over how they work, with performance judged by output and original expert judgment. The role may begin on the team best suited to the hire’s expertise, with opportunities to move across teams and potentially advance quickly into senior roles. The position is remote, hybrid, or onsite in San Francisco; onsite work is preferred. Commitment is full-time on the worker’s own schedule, with most team members working 40 or more hours per week and a few pod check-ins each week. The position is a 1099 contract (or W-8BEN internationally) with no end date. Pay is guaranteed during weeks 1–4 while ramping, then is paid per certified task; most people in the position earn $120,000–$180,000 or more, with uncapped upside.

Requirements

PhD in cell biology, bioimage analysis, pathology, or a related field, or an MD with pathology training.
Senior, hands-on imaging experience, including experience examining thousands of samples and recognizing normal and abnormal findings.
Direct experience with image segmentation tools such as CellProfiler, Cellpose, StarDist, QuPath, ilastik, napari, or Fiji.
Deep expertise in at least one of bioimage analysis, cell biology, or histopathology, and working fluency across the others.
Strong command of experimental design, image quantification, and statistics.
Daily use of Claude, Cursor, or similar AI agents, with the judgment to identify model reasoning that is technically correct but scientifically unsound.
Ability to overrule an AI agent when evidence demands it and openness to updating one’s view when shown wrong.
Ability to calibrate confidence and communicate uncertainty rather than defaulting to binary calls.
Must meet the stated work authorization criteria: US citizen, permanent resident, or OPT holder; STEM OPT extensions and sponsorship are not supported.

Nice To Haves

Electron microscopy experience.
Pathology experience.
Pharmacology experience, such as pharmacodynamic or drug-uptake imaging assays.
Familiarity with public resources such as the Image Data Resource (IDR), BioImage Archive, Human Protein Atlas, Cell Painting Gallery, or TCGA whole-slide images.

Responsibilities

Define expert reasoning for imaging questions and write gold-standard answers with defensible rationales.
Design realistic scenarios covering sample preparation, staining, acquisition, preprocessing, segmentation, quantification, and interpretation.
Curate image datasets across immunofluorescence, confocal and live-cell imaging, electron microscopy, histology and histopathology, and high-content screening.
Build Python analysis pipelines, including deep-learning segmentation, to produce verifiable ground truth.
Partner with software engineers to turn scientific judgment into new workflows.
Test whether agents distinguish true signal from artifacts such as bleed-through, uneven illumination, fixation damage, out-of-focus planes, and segmentation failure.
Adjudicate ambiguous images, identify what the data supports and where experts may disagree, and determine what additional evidence could resolve a case.
Assess whether agents select appropriate imaging modalities, resolutions, and units of analysis for scientific questions.
Interrogate agent reasoning step by step and challenge conclusions that would not withstand expert review.
Identify where agents’ reasoning fails, what an expert would do instead, and what evidence would change the conclusion.
Score agent confidence, rewarding explicit uncertainty rather than overconfident conclusions on ambiguous images.