Applied Scientist — Agentic AI for Scalable GRC

Amazon Inc.; Seattle (WA); Presencial; USD 143.000 - 193.000; Tempo integral

Resumo da vaga

Amazon Science is hiring to own and optimize the end-to-end NLI model used for governance, risk, and compliance decisions. You will fine-tune, calibrate thresholds, and monitor model drift across multiple frameworks in a production setting.

You will design decomposition pipelines that turn compliance controls into NLI hypotheses, improve evidence linking with ranking and embeddings, and tackle OCR quality challenges while preserving privacy and data safeguards.

Qualificações

• 2+ years of building models for business application.
• 2+ years of programming in Java, C++, Python or related language.
• 2+ years of hands-on predictive modeling and large data analysis experience.

Responsabilidades

• Own the production NLI model end-to-end: fine-tune, promote, calibrate thresholds, and monitor for drift to ensure accurate compliance verdicts across frameworks.
• Design and implement framework decomposition pipelines that translate compliance controls into NLI hypotheses, including enrichment and per-framework threshold calibration.
• Build and improve the evidence auto-linking system using ranking, embedding, and reranking techniques, tuning precision-recall tradeoffs and addressing OCR quality challenges.
• Drive the quality flywheel by establishing labeling standards, designing offline and online evaluation methods, orchestrating retraining cycles, and ensuring privacy-safe data handling.
• Research and develop the agent-context capability — retrieval and representation of lifecycle signals, cross-framework knowledge, memory, grounding, and hallucination measurement — collaborating with engineering pods to bring these models into production.

Conhecimentos

Java

C++

Python

Predictive modeling

Large data analysis

Formação

PhD in computer science, ML, engineering, or related fields

Descrição da vaga

Amazon Science is hiring to own and optimize the end-to-end NLI model used for governance, risk, and compliance decisions. You will fine-tune, calibrate thresholds, and monitor model drift across multiple frameworks in a production setting.

You will design decomposition pipelines that turn compliance controls into NLI hypotheses, improve evidence linking with ranking and embeddings, and tackle OCR quality challenges while preserving privacy and data safeguards.