New opportunity

AI Scientist

Nabla Bio · Boston, United States

About this role

Nabla Bio is hiring an AI Scientist to lead development of foundation models for biomolecular generation and prediction in therapeutic design. The models are evaluated using large-scale experimental feedback, including wet-lab assays of therapeutic function, safety, and precision. The posting lists Boston, MA, USA, as its location and specifies that the role is in person in Cambridge, MA.

Responsibilities

Design, implement, and evaluate training data, model architectures, training schemes, and loss functions for biomolecular generation and prediction.
Improve generative and predictive performance using experimental results. The platform can test dozens of modeling hypotheses in parallel and return results across a million drug designs within weeks.
Work with AI engineers to productionize models for internal and pharmaceutical-partner design workflows.
Advance approaches in machine learning, protein modeling, and sequence design, working closely with experimentalists to connect model behavior to biological outcomes.

Qualifications

At least 5 years of experience developing deep learning models, with strong Python and PyTorch engineering skills.
Experience with distributed training and scaling large models in HPC or cloud environments; a record of creativity, rigor, and technical leadership in ML research; and comfort collaborating with experimentalists.
Prior experience in generative modeling, protein or biomolecular ML, or large-scale sequence modeling is a plus.

The full-time role offers collaboration with wet-lab biology, bioinformatics, and software engineering experts, plus a competitive salary, equity, and benefits package. No salary amount is stated.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

Deep LearningMachine Learning ResearchFoundation ModelsGenerative ModelingProtein ModelingBiomolecular ModelingSequence DesignPythonPyTorchDistributed TrainingHigh-Performance ComputingCloud ComputingModel TrainingExperimental ValidationTechnical Leadership

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Lead foundation-model development
The role owns core biomolecular models for generation and prediction, including training data, architectures, training schemes, and loss functions. Prepare to explain how you would compare modeling choices and evaluate progress.
Improve models using experimental feedback
The posting emphasizes wet-lab assays of therapeutic function, safety, and precision—not just loss curves and benchmarks. Prepare to describe how assay results could inform the next modeling iteration.
Scale rigorous ML research
The role calls for technical leadership and experience with distributed training of large models in HPC or cloud environments. Prepare to discuss how you would make large-scale experiments reproducible and interpretable.
Move research into design workflows
The scientist will work with AI engineers to productionize models for internal and pharmaceutical-partner workflows. Prepare to explain how you would hand off and validate a research model for practical use.

Likely challenges to think through

Reconciling model metrics with biological outcomes
A likely challenge is that improvements on computational metrics may not translate into better assay results. Prepare to discuss how you would investigate disagreement and choose what to optimize next; this is a preparation hypothesis, not a reported problem.
Choosing experiments at platform scale
The posting says the platform can test dozens of hypotheses in parallel and return results across a million drug designs within weeks. A likely trade-off is breadth versus interpretable comparisons. Prepare a principled way to prioritize hypotheses and learn from the results.
Balancing research novelty with deployability
The role spans advancing modeling approaches and productionizing them with AI engineers. A likely difficulty is weighing performance gains against training cost and workflow reliability. Prepare to explain how you would evaluate that trade-off without assuming a particular company process.

Practice your answers

Technical practice

How would you design training data for a biomolecular model used for both generation and prediction?

What to coverPreparation suggestion: Discuss data quality, representation, coverage, and how you would prevent misleading evaluation from closely related examples. Relate those choices to the role’s stated responsibility for designing and evaluating training data.

How would you compare model architectures for biomolecular generation and prediction?

What to coverPreparation suggestion: Explain how you would compare candidate architectures on predictive utility, quality of generated designs, and training cost. The posting calls for architecture design, but does not specify an architecture to favor.

How would you choose training schemes and loss functions for these modeling goals?

What to coverPreparation suggestion: Describe how objectives could balance generation and prediction, and how you would detect a loss improvement that does not translate into useful designs. The role explicitly includes evaluating training schemes and loss functions.

How would you evaluate a model when wet-lab results differ from offline benchmarks?

What to coverPreparation suggestion: Separate evidence from offline metrics and assays, then investigate whether data coverage, measurement variation, or the chosen objective explains the gap. Ground your answer in the posting’s emphasis on therapeutic function, safety, and precision alongside loss curves and benchmarks.

How would you use experimental feedback to improve generative and predictive performance?

What to coverPreparation suggestion: Explain how you would analyze which designs succeeded or failed, form a testable modeling hypothesis, and evaluate a change against prior results. The role specifically calls for performance improvements based on experimental feedback.

How would you compare many modeling hypotheses when experimental results arrive at large scale?

What to coverPreparation suggestion: Describe a comparison that isolates meaningful changes, accounts for uncertainty, and prioritizes follow-up work. The posting says the platform can test dozens of hypotheses in parallel and return results across a million drug designs within weeks; it does not prescribe a testing protocol.

How would you scale training of a large model across HPC or cloud resources?

What to coverPreparation suggestion: Cover distributed-training choices, throughput and memory trade-offs, and checks that training remains correct as it scales. Distributed training and scaling large models in HPC or cloud environments are stated qualifications.

How would you troubleshoot an unstable PyTorch training run?

What to coverPreparation suggestion: Walk through checks on inputs, gradients, numerical behavior, and reproducibility before changing the model. Python and PyTorch engineering fluency is required; this debugging scenario is a suggested way to practice demonstrating it.

What would you consider when moving a research model into a design workflow?

What to coverPreparation suggestion: Discuss dependable inference, validation of model outputs, and how you would work with AI engineers to preserve research behavior in use. The posting specifies productionizing models for internal and pharmaceutical-partner design workflows.

How would you decide whether a new protein-modeling or sequence-design approach is worth pursuing?

What to coverPreparation suggestion: Compare its relevant evidence, implementation cost, and potential to improve biological outcomes against a strong baseline. The role calls for advancing the state of the art in ML, protein modeling, and sequence design; it does not name a required method.

Behavioural practice

Tell me about a time you led an ML research direction through uncertainty.

What to coverPreparation suggestion: Give the situation, the decision and actions you led, and the result, including how you judged the evidence. This practices the creativity, rigor, and technical leadership named in the qualifications.

Tell me about a time experimental evidence challenged your model’s apparent performance.

What to coverPreparation suggestion: Describe the situation, how you investigated the discrepancy with relevant collaborators, what you changed, and the result. This reflects the role’s emphasis on connecting model behavior to biological outcomes rather than relying only on benchmarks.

Tell me about a time you worked with engineers to put a research model into use.

What to coverPreparation suggestion: Explain the workflow need, your actions during the handoff, and the resulting outcome or lesson. The posting calls for collaboration with AI engineers to productionize models for design workflows.

Tell me about a time you collaborated with specialists outside ML on a difficult problem.

What to coverPreparation suggestion: Set out the situation, how you reconciled different technical perspectives, and what resulted. The role involves close work with experimentalists and collaboration across wet-lab biology, bioinformatics, and software engineering.

Tell me about a time you had to prioritize competing research hypotheses.

What to coverPreparation suggestion: Describe the constraints, how you selected what to test, and what the results taught you. This is suggested practice based on the posting’s scope for testing multiple modeling hypotheses; it is not a claim about a specific team process.

Sharpen your knowledge

Topics to prepare

Biomolecular model design
The role designs training data, architectures, training schemes, and loss functions for generation and prediction. As preparation, sketch two modeling approaches for a hypothetical biomolecular sequence task and explain how you would compare their objectives and outputs.
Experimental validation
The posting emphasizes wet-lab assays of therapeutic function, safety, and precision alongside model benchmarks. As preparation, outline how you would interpret a case where offline performance improves but assay results worsen, and what evidence you would seek before changing the model.
Learning from large-scale feedback
The role calls for improving generative and predictive performance using experimental results across many designs and parallel hypotheses. As preparation, design a comparison that separates a genuine modeling improvement from variation in the tested designs.
Scaling deep learning research
Experience with distributed training and scaling large models in HPC or cloud environments is required. As preparation, explain the trade-offs you would examine when increasing model size, training data volume, and the number of training workers.
Protein and sequence modeling
The role advances protein modeling and sequence design; prior protein or biomolecular ML or large-scale sequence modeling is listed as a plus. As preparation, compare a generation objective with a property-prediction objective for hypothetical protein sequences, noting what each would need experimental validation to establish.

Skills to revise

Python and PyTorch model implementation
Strong Python and PyTorch engineering fluency is required. As preparation, implement a small sequence generation or prediction model with a configurable loss, then run a brief training and evaluation check.
Distributed training diagnosis
The posting requires distributed-training experience in HPC or cloud environments. As preparation, work through a small multi-worker training example and document how you would check correctness, throughput, and differences from a single-worker run.
Assay-feedback analysis
The role uses experimental results to improve models and connects model behavior to biological outcomes with experimentalists. As preparation, analyze a hypothetical table of predictions and function, safety, and precision assay results; identify one follow-up modeling hypothesis and questions for the experimental team.
Research-to-production handoff
The scientist collaborates with AI engineers to productionize models for internal and pharmaceutical-partner design workflows. As preparation, draft a concise handoff for a hypothetical model covering its inputs, outputs, evaluation evidence, limitations, and checks needed before use in a design workflow.

Your preparation checklist

Prepare a model-development walkthrough
The role covers training data, architectures, training schemes, and loss functions. Prepare one relevant project walkthrough showing your design choices, baselines, evaluation results, and what you changed after testing.
Practice a biomolecular modeling design exercise
Sketch a proposed generation-and-prediction workflow, including data checks, model and loss choices, and measures of generative and predictive performance. Treat it as a discussion exercise, not a description of Nabla Bio’s existing system.
Explain how you would use experimental feedback
The posting emphasizes wet-lab assays of therapeutic function, safety, and precision. Prepare a plan for comparing modeling hypotheses and using assay results to guide the next training or design cycle.
Gather engineering and scaling evidence
Prepare concrete examples of Python and PyTorch engineering and distributed training in HPC or cloud environments, where you have them. Note the scale, your decisions, bottlenecks, and measured outcomes.
Show research leadership and collaboration
Select an example that demonstrates rigorous ML research and technical leadership. Be ready to explain how you worked with experimentalists to connect model behavior to biological outcomes, if applicable.
Prepare a productionization discussion
The role involves working with AI engineers on internal and pharmaceutical-partner design workflows. Outline how you would evaluate model readiness, hand off a reproducible implementation, and monitor performance after deployment.
Write targeted questions for the team
Ask how modeling hypotheses are prioritized, how assay feedback reaches model teams, what productionization entails, and how the AI Scientist collaborates with wet-lab biology, bioinformatics, and engineering colleagues.

Availability last checked: Nov 4, 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in United States. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…