Visa Research is looking to conduct world-class research on data analytics and contribute to the long-term research agenda in large-scale data analytics and machine learning, as well as deliver innovative technologies and insights to Visa's strategic products and business. This role represents an exciting opportunity to make key contributions to Visa's strategic vision as a world-leading data-driven company.
Requirements
- Proficiency with object oriented programming languages, such as Python, C++, Java, and Lua.
- Experience with multi-agent systems and frameworks
Responsibilities
- Research and develop methods to reduce hallucinations and improve factual reliability in AI agents.
- Design access control mechanisms and policy enforcement for secure AI agent interactions.
- Investigate and implement attack mitigation strategies, including Prompt injection defenses and Jailbreak prevention methods
- Explore multi-agent architectures for secure, coordinated, and efficient collaboration between AI agents.
- Build and evaluate efficiency improvements such as model compression, quantization, and optimized LLM fine-tuning.
- Develop benchmarking frameworks for safety, robustness, and performance evaluation.
- Contribute to top-tier research publications
Other
- Students pursuing a Ph.D. with a graduation date in December 2026 or later
- Strong communications skills, specifically, the absence of repeated grammatical or typographical errors, clear and concise written and spoken communications that demonstrate professional judgment.
- Strong track record in research and impact in the research community in highly reputed journals and conferences through multiple publications, patents and/or presentations at conferences.
- Proven research experience in AI safety, security, or alignment
- Strong verbal and written communication skills and an eagerness to collaborate with members of your team as well as other Visa teams
- Demonstrated ability to think outside the box and innovate
- The ability to take on challenges and address problems head-on
