29 days ago
Berkeley, CA, USA or San Francisco, CA, USAIntern
H1B sponsor
Responsibilities
- Design and review APIFlow-Bench tasks and mock API environments while extending its Python evaluation harness and task-generation pipeline.
- Build an action-level AI safety benchmark covering prompt injection, data exfiltration, permission overreach, and auditable evaluation.
- Fine-tune open-weight models for tool calling and agentic tasks using supervised fine-tuning, distillation, reinforcement learning, and PyTorch.
- Run benchmarked experiments, ablation studies, error analyses, and cost-aware reporting of training results.
- Evaluate ultra-low-bit quantized models against full-precision models for on-device use.
- Help build Postman’s Agent Mode in TypeScript using tool loops, multi-step execution, and checkpointing.
- Document experiments, design decisions, and runbooks, and identify safety, fairness, and privacy concerns.
Requirements
- Currently pursuing a BS, MS, or PhD in Computer Science, Data Science, or a related quantitative field.
- Hands-on experience training or evaluating machine-learning models through coursework, research, hackathons, or internships.
- Solid Python fundamentals, including data structures, functions, basic testing, and writing code outside notebooks.
- Working knowledge of at least one deep-learning framework, preferably PyTorch.
- Clear written and verbal communication and a consistent practice of documenting technical work.
- Preferred qualifications include experience with open-weight LLM fine-tuning, LLM agents, evaluation harnesses, statistical reporting, APIs and services, CLIs, Docker, cloud, AI safety, model quantization, technical publications, side projects, or AI coding tools.
Benefits
- Summer 2026 internship with flexible scheduling and mentorship from senior engineers.
- Full medical coverage, flexible PTO, wellness reimbursement, monthly lunch stipend, team-building events, and donation matching.
- In-office collaboration five days per week in San Francisco Bay Area, Boston, Austin, New York City, Tokyo, and London; Bangalore is three days per week currently and is expected to transition to five days per week by year-end.
Tech Stack
Categories
About Postman
Postman builds a SaaS platform for designing, testing, documenting, and managing APIs used by software teams across development and operations. The company sells free and paid plans, including enterprise features for collaboration, security, and governance, and offers desktop and cloud tools. Founded in 2014 and headquartered in San Francisco, Postman is privately held and reports adoption by over 40 million developers and 500,000 organizations worldwide.