Research Intern

ABOUT US Sierra is the leading platform for customer-facing AI agents, working with many of the world's biggest brands — including The GAP, Rocket Mortgage, SoFi, Sutter Health, and SoftBank — to transform how they serve customers and grow their businesses. We are primarily an in-person company based in San Francisco, with growing offices across North America, Europe, and Asia. We are guided by a set of values that are at the core of our actions and define our culture: Trust, Customer Obsession, Craftsmanship, Intensity, and a commitment to balancing Family along the way. These values are the foundation of our work, and we are committed to upholding them in everything we do. Our co-founders are Bret Taylor https://www.linkedin.com/in/brettaylor/ and Clay Bavor https://www.linkedin.com/in/claybavor/. Bret currently serves as Board Chair of OpenAI. Previously, he was co-CEO of Salesforce (which had acquired the company he founded, Quip) and CTO of Facebook. Bret was also one of Google's earliest product managers and co-creator of Google Maps. Before founding Sierra, Clay spent 18 years at Google, where he most recently led Google Labs. Earlier, he started and led Google’s AR/VR effort, Project Starline, and Google Lens. Before that, Clay led the product and design teams for Google Workspace. Sierra’s Applied Research team works to make the models behind our agents better, faster, and cheaper across many kinds of tasks. Our work covers post-training open-weight models for specialized capabilities, new approaches to voice AI, and the benchmarks the field uses to measure agents, including τ-bench. As an Applied Research Intern, you’ll join the team for a research project, working with a mentor, you’ll pick a problem that matters to how AI agents perform in the real world, run rigorous experiments, and potentially write up what you find for a top venue. If you want your research to reach both the research community and production agents used by millions of people, come do it at Sierra. WHAT YOU’LL DO - Own a research project from start to finish, from framing the question and running experiments to writing and submitting a paper. - Push agent evaluation forward, for example by extending τ-bench to new domains, modalities, or long-horizon tasks, or by building new ways to measure reliability and consistency. - Study how agents reason, use tools, and follow policies over multi-turn conversations, and find out where and why they fail. - Explore post-training methods such as supervised fine-tuning and reinforcement learning to improve agent behavior, and measure the gains rigorously. - Work with large-scale, proprietary datasets of text and audio interactions to find research questions grounded in real customer problems. - Work with researchers and engineers across Sierra, and where it makes sense, turn your findings into improvements in production. WHAT YOU’LL BRING - Currently pursuing a PhD in computer science, machine learning, or a related field. - A research track record, such as first-author publications or submissions at venues like NeurIPS, ICML, ICLR, ACL, EMNLP, Interspeech or ICASSP. - Strong experimental skills: you design clean experiments, build careful baselines, and are honest about what the results do and do not show. - Fluency in Python and modern ML tooling (for example, PyTorch, and LLM inference and training stacks). - High agency and a builder’s mindset. You’ll work across the stack to get an experiment running. - Clear, direct communication, both in writing and when presenting research. EVEN BETTER - Research experience with LLM agents, tool use, or multi-turn dialogue. - Experience designing benchmarks or evaluations, especially ones tied to real-world outcomes. - Experienc