Contribute to APIFlow-Bench, design and review benchmark tasks and their mock API environments, extend the evaluation harness and task-generation pipeline in Python, and help maintain the public multi-model leaderboard with statistical confidence intervals.
Help build a new action-level AI safety benchmark, including scenario design, threat modeling (prompt injection, data exfiltration, permission overreach), and auditable evaluation design.
Fine-tune open-weight models for tool calling and agentic tasks (SFT, distillation, and RL) using PyTorch and the open-source training ecosystem, on both managed training platforms and self-managed cloud GPUs.
Design and run experiments with rigor: evaluate every training run on benchmarks, support ablation studies and error analysis, track experiments, and report results honestly, including cost.
Evaluate ultra-low-bit quantized models for on-device use: extend quantized vs. full-precision benchmark comparisons and analyze divergences.
Requirements
BS, MS, or PhD in Computer Science, Data Science, or a related quantitative field
Nice to have: Fine-tuning open-weight LLMs (SFT, LoRA, RL, or distillation)
Nice to have: Building LLM agents (tool calling, multi-step loops) or LLM evaluation harnesses/benchmarks
Nice to have: Shipping real software end-to-end (APIs, services, CLIs, Docker, CI/CD, cloud)
Nice to have: AI safety and robustness (red-teaming, prompt injection, agent security, fairness, or interpretability)
Staff Engineer to join the Fabric Gateway team, shaping the architecture of the data plane, control plane, registry, and policy engine for Postman's protocol-agnostic control plane for the agentic world.
Locations (7)
San Francisco, California, United States
Boston, Massachusetts, United States
New York, New York, United States
Austin, Texas, United States
Tokyo, Tokyo, Japan
London, England, United Kingdom
Bangalore, India
Responsibilities
Drive the technical vision and architecture for Fabric Gateway across its registry, control plane, data plane, and SDKs and connectors.
Design and build a high-performance, horizontally scalable data plane that routes HTTP, gRPC, MCP, and A2A traffic with minimal added latency.
Build the policy engine that authenticates, authorizes, classifies, and logs every agent interaction before it happens, with fail-closed behavior when policy cannot be verified.
Own reliability features on the hot path: fallback chains, circuit breakers, health-aware routing, rate limiting, and weighted traffic distribution across LLM and API providers.
Design secure credential brokering so upstream secrets are injected by the gateway and never exposed to agents.