AI Engineer, Agents for Biochemistry

Responsibilities

1. Build agent harnesses and measure them against each other: memory strategies, context budgeting, tool interface design, and the protocols by which agents hand work to one another
2. Debug agent failures at the trajectory level and resolve them in the harness
3. Build and maintain the retrieval layer between the agents and the knowledge graph, including query planning, schema-aware retrieval, and reranking
4. Implement schema validation, provenance tracking, and rejection handling at the write boundary
5. Implement inference methods for incomplete and conflicting evidence, including belief revision, uncertainty propagation, and abstention
6. Build extraction pipelines that convert publications, patents, and instrument output into structured records
7. Integrate third-party scientific databases and reconcile conflicts between sources
8. Construct benchmarks for scientific reasoning tasks, including task selection, labeling strategy, and scoring methodology
9. Own the automated evaluation harness used to gate model and prompt changes
10. Work directly with pilot users to identify failure modes and translate them into system changes

Required qualifications

• Demonstrated work building multi-step language model agents that have run under real users OR in applying LLMs to perform complex tasks
• Experience designing evaluations for LLM systems, including benchmark construction and model-based grading
• Strong Python, with experience building and operating production backend services
• Experience modeling a complex domain in a typed schema, relational database, or graph database
• Experience with retrieval systems beyond embedding similarity search
• Ability to read primary literature in biology or chemistry, or clear interest in developing it

Preferred qualifications

• Experience with multi-agent orchestration and MCP
• Experience with probabilistic programming, Bayesian inference, or calibration measurement
• Experience with non-monotonic logic, truth maintenance, or other defeasible reasoning systems
• Experience with information extraction from unstructured documents, including table and layout parsing
• Experience with reinforcement learning or finetuning of language models
• Background in cheminformatics, including RDKit, SMILES normalization, or optical chemical structure recognition
• Background in chemical biology, targeted protein degradation, or mass spectrometry proteomics
• Experience with scientific data infrastructure such as ELN, LIMS, or FAIR data standards