技术员工成员（软件工程师，连接器平台）
Member of Technical Staff (Software Engineer, Connector Platform)
关于该职位
Connector Platform 团队构建了让 Perplexity 的代理访问世界软件的数据层。该团队负责将数百种异构集成（原生、MCP、CLI、第一方和第三方 API）转化为一个统一、可靠、类型良好的表面，代理可以放心地调用。
Connector 平台是 Computer 的核心层，构成了知识层：它是代理发现可用工具、理解每个工具的含义、决定调用哪个工具，并基于真实、授权、最新的企业数据进行推理的方式。我们在连接器之上维护一个知识层，将上下文推送到连接器中，而不是让每个连接器自行保存组织知识，使 Computer 成为机构知识的权威来源。模型正在商品化；对客户真实系统的接地、可操作、授权的访问则不是。当这一层快速、准确且语义丰富时，建立在它之上的每个代理都会变得更智能；当它薄弱时，无论模型质量如何都无法弥补。
主要职责
- 负责连接器运行时的设计与实现，该系统注册、托管并通过单一代理接口执行内置连接器、托管的 MCP 服务器和 CLI 支持的工具。
- 构建和扩展语义层：工具和实体模式、能力元数据、关系建模，以及捕捉和应用组织和账户特定修正和知识的机制。
- 设计代理使用的工具发现和工具选择界面，优化模型准确性和上下文效率。
- 使代理循环更加稳健：结构化结果、部分失败和重试语义、幂等性、分页、速率限制处理，以及每个代理调用的可观测性。
- 定义连接器的认证、授权和凭证隔离模式（OAuth 流程、BYOK、按组织凭证边界），与安全和后端平台合作实现纵深防御。
- 构建连接器上架流程（模式、fixture 和评估套件），使新连接器以可衡量的质量发布，而非依赖运气，并推动评估指标以确认连接器在代理循环中实际有效。
- 设定连接器可靠性与可操作性的技术标准：SLA、可观测性、错误率监控和事件响应。
查看英文原文
ABOUT THE ROLE
The Connector Platform team builds the data layer that lets Perplexity's agents reach into the world's software. This team owns the systems that turn hundreds of heterogeneous integrations (native, MCP, CLI, first-party, and third-party APIs) into one unified, reliable, well-typed surface that agents can call with confidence.
The connector platform is the core layer that forms the knowledge layer for Computer: it is how the agent discovers what tools exist, understands what each one means, decides which to call, and grounds its reasoning in real, permissioned, up-to-date enterprise data. We maintain a knowledge layer above connectors that pushes and pulls context into them, rather than letting each connector hoard org knowledge on its own, making Computer the source of truth for institutional knowledge. Models are commoditizing; grounded, actionable, permissioned access to a customer's real systems is not. When this layer is fast, accurate, and semantically rich, every agent built on top of it gets smarter; when it is weak, no amount of model quality compensates.
KEY RESPONSIBILITIES
- Own the design and implementation of the connector runtime, the system that registers, hosts, and executes built-in connectors, hosted MCP servers, and CLI-backed tools behind a single agent-facing interface.
- Build and extend the semantic layer: tool and entity schemas, capability metadata, relationship modeling, and the mechanisms for capturing and applying organization- and account-specific corrections and knowledge.
- Design the tool-discovery and tool-selection surfaces that agents use to find the right connector and call it correctly, optimizing for both model accuracy and context efficiency.
- Make agent loops robust: structured results, partial-failure and retry semantics, idempotency, pagination, rate-limit handling, and observability into every tool call an agent makes.
- Define authentication, authorization, and credential-isolation patterns for connectors (OAuth flows, BYOK, per-org credential boundaries), partnering with Security and Backend Platform on defense-in-depth.
- Build the connector onboarding path (schemas, fixtures, and evaluation suites) so new connectors ship with measurable quality rather than hope, and drive the eval metrics that tell us a connector actually works inside agent loops.
- Set the technical bar for connector reliability and operability: SLAs, observability, error-rate monitoring, and incident response for an always-on, high-fan-out integration surface.
- Partner with product and AI teams to define clear connector interfaces and integration patterns so new agent capabilities can reliably build on the shared platform.
QUALIFICATIONS
- Experience designing and building backend systems that run in production (typically 4+ years for mid-level, more for senior and staff).
- Strong system design skills, with a track record of building efficient, reliable, and scalable architectures, ideally including API integration, gateway, or platform-style systems with many heterogeneous downstreams.
- Strong proficiency in at least one backend language such as Python, Go, or Rust, and the ability to work effectively in a multi-language environment.
- Hands-on experience with modern infrastructure (for example AWS, Kubernetes, and related cloud technologies).
- Depth in at least one of: OAuth and authorization protocols, API/connector or MCP-server development, schema and semantic modeling, or building tooling and evaluation for LLM-based agents.
- Comfort working in security-sensitive areas (auth, authorization, credential isolation) and making pragmatic trade-offs between safety, simplicity, and velocity.
- Collaborative mindset and eagerness to solve hard, ambiguous problems alongside other experienced engineers.
If you’re excited about this role, we encourage you to apply even if your experience doesn’t match every qualification listed above.