AI Researcher

About the Role

Talus Bio is looking for an AI Researcher to join the team full-time in our Seattle office (this is an in-person role). This person will be naturally curious, results-driven, and eager to work in a small preclinical biotech environment. You will improve Ptarmigan, our structure-free model that predicts which small molecules engage which proteins across the human proteome, and you will build new models for our drug discovery programs and partners.

Ptarmigan learns from Talus Bio's own chemoproteomics data: hundreds of millions of compound-protein measurements collected in native human cells, including on transcription factors and other proteins without good structures. We measure success by delivering value beyond in silico metrics: successful models drive early discovery and accelerate the optimization of molecules moving toward the clinic. You will work closely with our team of multi-disciplinary scientists to choose what models, metrics, and new features maximize our success. You will also work with our Product and Applications Lead to understand what partners need from our models.

This role suits a machine learning researcher who has applied models to drug discovery, ideally in a preclinical program, and wants to shape the modeling direction at a small company. We will set the level (AI Researcher or Senior AI Researcher) based on experience.

What you will do

• Improve Ptarmigan: Develop new architectures, training objectives, and data representations that improve how well Ptarmigan predicts compound-protein engagement.
• Build new models: Develop protein-ligand interaction and generative chemistry models that support hit discovery, hit-to-lead campaigns, and lead optimization, particularly for targets that lack structures or known binding pockets.
• Evaluate rigorously: Design benchmarks, held-out splits, and baselines (e.g., docking and structure-based co-folding models) that reflect how the models will be used in practice, and report where they fail.
• Close the loop with the lab: Work with our experimental scientists to pick compounds and experiments for prospective validation and active learning, so that each round of data improves the next model.
• Build agentic tools: Build tools and agents that call our models through APIs and MCP, so that scientists and AI agents can run virtual screens and interpret the results.
• Ship your work: Write tested, version-controlled code and take models from research to production, working with our team on training and inference infrastructure in the cloud.
• Share results: Present plans and results at R&D meetings and to partners, and publish papers, preprints, and open-source software where it advances our science.

Qualifications

• Ph.D. in machine learning, computer science, computational chemistry, bioinformatics, or a related field with 2+ years of industry or postdoctoral experience, or an M.S. with 5+ years of industry experience
• Experience applying machine learning to small molecules or protein-ligand interactions in a drug discovery setting
• Experience designing and training deep learning models on molecular or protein data with PyTorch or a similar framework
• Fluency in Python and Git, and experience training models on GPUs in the cloud
• Ability to explain models, their results, and their limitations to chemists, biologists, and other non-computational colleagues
• Experience building software and managing projects using agentic AI tools, such as Claude Code or OpenAI Codex
• Able to work onsite in Seattle

Preferred

• Experience building models that were used in a preclinical program for hit discovery, hit-to-lead, or lead optimization
• Experience with generative models for small molecules (e.g., diffusion, flow matching, or chemical language models)
• Experience with building protein language models or structure-based models (e.g., Boltz, AlphaFold, or docking)
• Experience with proteomics, mass spectrometry, or chemoproteomics data, including activity-based protein profiling
• Experience building custom LLM agents or tools that use the Model Context Protocol (MCP)
• A record of publications or open-source software in machine learning for biology or chemistry
• Experience with active learning, uncertainty estimation, or Bayesian optimization
• Experience leading ML projects or mentoring other scientists