Machine Learning Researcher - Agentic Science

Role Overview

In this role, you will develop the agentic research vertical at PostEra, using agentic systems to automate the development of mechanistic models for biochemical and physiological processes, and analyse biological data for target validation. You will interact closely with chemists and biologists to use these models to drive drug discovery decisions. 

You will also develop machine learning methods that can rapidly adapt to new drug discovery problems from limited labeled data. A particular focus is molecular and tabular in-context learning and building foundation models from PostEra’s proprietary multimodal data. You will build models that use the context of drug discovery effectively, determine which prior examples and tasks are relevant, quantify when transfer is helpful or harmful, and provide reliable predictions under distribution shift. 

You will help drive the full research loop: defining tasks, constructing datasets and evaluation episodes, developing strong baselines, training and scaling models, performing rigorous ablations, and translating successful methods into capabilities used by PostEra’s scientists. Prior drug discovery experience is not required, but you should be motivated to learn the domain and work closely with medicinal chemists, computational chemists, and other scientists.

Key Responsibilities

• Agentic Research for Chemistry and Biology: Develop and benchmark agentic systems to automate the development of quantitative models to simulate biological and physiological processes, and the analysis of biological data. 

• Research Direction and Execution: Independently identify, formulate, and lead research projects involving in-context learning, agentic systems, few-shot adaptation, tabular foundation models, and molecular machine learning.

• In-Context Learning for Molecules: Design and train models that adapt to new assays, endpoints, targets, or chemical series using limited labeled context and heterogeneous historical data.

• Benchmarking: Rigorously compare new approaches against strong baselines, curating test cases that deconfounds impact of different sources of bias. 

• Cross-Functional Collaboration: Work with scientists to connect modeling objectives and evaluation metrics to practical decisions in potency modeling, ADME prediction, selectivity, lead optimization, and the design of early clinical studies.

• Model Training and Scaling: Develop efficient training and data pipelines and, where appropriate, scale models across large collections of molecular and tabular tasks.

• Research Engineering: Produce readable, reproducible research code; maintain well-tracked experiments; and contribute through code review, documentation, and shared modeling infrastructure.

• Scientific Dissemination: Publish results in leading machine learning, medicinal chemistry, or computational biology venues. You will be an ambassador of PostEra to the scientific community. 

Key Requirements

• PhD degree in machine learning, or STEM research involving the development of novel machine learning approaches 

• Track record of high-quality research, such as publications and open-source contributions

• Strong research or engineering experience in modern machine learning, deep learning, or statistical modeling, backed by understanding of the theory behind machine learning algorithms

• Demonstrated expertise in at least one relevant area: agentic workflows for science, machine learning approaches to bioinformatics and clinical data modelling, in-context learning, tabular learning, few-shot learning

• Hands-on experience training, debugging, and evaluating ML models in Python using frameworks such as PyTorch or JAX

• Ability to independently translate ambiguous scientific or technical problems into well-defined ML projects, including datasets, task definitions, baselines, metrics, and validation schemes

• Ability to design careful experiments, benchmarks, and ablations that distinguish improvements from biases, and understand which aspects of the model led to the improvements

• Comfort working in a startup environment where priorities evolve, data is imperfect, and high-quality judgment matters as much as raw model complexity

Nice to Haves

Candidates are more likely to succeed in this role if they also have experience with one or more of the following:

• Training tabular foundation models, particularly for sparse, heterogeneous, small-data, or high-missingness settings

• Developing molecular in-context learning systems or adapting general-purpose in-context models to molecular or scientific data

• Large model training, including 1B+ parameter models, distributed training, sharding, data parallelism, model parallelism, and large-scale data pipelines

• The development of AI “co-scientist” systems for physical or biological problems 

• Hands-on experience in modelling biological, biochemical or clinical data using machine learning approaches 

• Moving research models into production scientific software or computational workflows