Applied Scientist: LLM Agents & Multimodal AI

Amazon Inc.; Piemonte; Hybrid; EUR 17,000 - 34,000; Full time

Job summary

Amazon's Kindle Direct Publishing team in Bengaluru is seeking a science-focused engineer to advance author support using large language models. You will help turn ideas into agentic solutions, orchestrating multiple agents and tools and applying multimodal models to improve catalog quality and discovery.

You will collaborate with global scientists to design offline and online experiments, post-train models with reinforcement learning, and deploy improvements that boost accuracy, cost

Qualifications

• PhD, or a Master’s degree and experience in CS, CE, ML or related field.
• 3+ years of building machine learning models for business applications.
• Experience programming in Java, C++, Python or related language.

Responsibilities

• Build the multi-agent system behind KDP’s author support assistant: orchestration across specialized agents and tools, personalization, and the evaluation that drives its continuous improvement.
• Build multimodal models of book attributes, and use them for similar-book search across languages, catalog quality, and category and keyword recommendations.
• Post-train open models on KDP expertise (for example with supervised fine-tuning and reinforcement learning), and deploy them where they beat general-purpose models on accuracy, cost, or speed.
• Work closely with our scientists in the US and with engineering partners to take your work to production, design rigorous offline and online experiments, and write clear readouts with recommendations.

Skills

Machine learning

Java

C++

Python

LLM-based agents

Prompt engineering

Education

PhD in CS/CE/ML

Master's in CS/CE/ML

Tools

Java

C++

Python

SQL

Job description

Amazon's Kindle Direct Publishing team in Bengaluru is seeking a science-focused engineer to advance author support using large language models. You will help turn ideas into agentic solutions, orchestrating multiple agents and tools and applying multimodal models to improve catalog quality and discovery.

You will collaborate with global scientists to design offline and online experiments, post-train models with reinforcement learning, and deploy improvements that boost accuracy, cost