Research Intern

Research Intern

Sierra builds customer-facing AI agents for some of the world's largest companies, including The GAP, Rocket Mortgage, SoFi, Sutter Health, and SoftBank. The company, founded by Bret Taylor and Clay Bavor, is headquartered in San Francisco with offices expanding across North America, Europe, and Asia. Sierra's leadership brings deep expertise from OpenAI, Salesforce, Facebook, and Google, where foundational technologies like Google Maps and Google Lens were created.

Sierra's Applied Research team focuses on making the models powering AI agents more capable, efficient, and cost-effective across diverse applications. The team works on post-training methods for open-weight models, novel approaches to voice AI, and industry benchmarks like tau-bench that measure agent performance. This internship offers the chance to conduct research that reaches both the academic community and production systems used by millions of people globally.

What you'll do

• Lead a research project independently, from problem definition and experimental design through paper writing and submission to a top-tier venue.

• Advance agent evaluation methodologies, such as expanding tau-bench into new domains, modalities, or long-horizon scenarios, or developing novel metrics for reliability and consistency.

• Investigate how agents reason, utilize tools, and maintain policies through multi-turn interactions, identifying failure points and root causes.

• Test post-training techniques including supervised fine-tuning and reinforcement learning to enhance agent performance, with rigorous measurement of improvements.

• Leverage large-scale proprietary datasets of text and audio interactions to surface research questions grounded in real customer needs.

• Collaborate with Sierra's research and engineering teams to translate findings into production enhancements.

What you'll bring

• Current enrollment in a PhD program in computer science, machine learning, or related discipline.

• Demonstrated research contributions, including first-author publications or accepted submissions to venues such as NeurIPS, ICML, ICLR, ACL, EMNLP, Interspeech, or ICASSP.

• Strong experimental methodology: you design rigorous experiments, establish solid baselines, and communicate results honestly and precisely.

• Proficiency in Python and modern machine learning frameworks such as PyTorch, along with experience in LLM inference and training infrastructure.

• Self-directed approach and hands-on builder mentality, willing to work across the technical stack to execute experiments.

• Effective communication skills, both written and verbal, for presenting research findings.

Nice to have

• Prior research involving LLM agents, tool use, or multi-turn dialogue systems.

• Experience developing benchmarks or evaluations connected to real-world performance.

• Background in supervised fine-tuning, reinforcement learning, or long-horizon optimization techniques.

• Familiarity with speech, voice, or multimodal machine learning models.

• Open-source contributions to machine learning research tools or benchmark repositories.

What they offer

• Flexible internship duration and start dates tailored to your academic calendar.

• Direct mentorship from Sierra's Applied Research team and embedded placement within the group.

• In-person work environment in San Francisco, with office presence Monday through Thursday.