AI Application Engineer (Agent Focus)

The company is building an Agent-based generative parametric 3D product, with core AI functions currently led directly by the founder. This role is not just about coding; it requires deep involvement in the engineering implementation from Demo to production environment, with a focus on building 'controllable, recoverable, observable' Agent workflows and tool systems. 1. Agent Orchestration & Stability · Responsible for Agent orchestration and execution mechanisms: shunting/routing, state management, idempotency design, and consistency strategies · Build 'anti-fragile' systems: error classification, replay/compensation, dead-letter queues, and recoverable links to ensure long-link interactions do not drop single steps or get stuck · Design multi-model routing strategies to dynamically select models based on task difficulty and budget, enabling Token cost visualization and budget control 2. Tools System & Evaluation · Design and implement reusable Agent Tools with clear input/output contracts (Schema Design), testable and canary releases · Build Agent offline playback and regression testing mechanisms, quantifying tool quality with metrics such as success rate, tool misuse rate, and hallucination rate 3. Observability & Reliability · Integrate Tracing/Metrics/Logging (e.g., OpenTelemetry) into core links, solving context propagation and issue localization across modules and services · Implement high-reliability consistency schemes (e.g., Transactional Outbox) as needed by business requirements Qualifications · Hands-on experience in LLM/Agent project implementation, with in-depth understanding of multi-turn dialogue state management, Context Engineering, tool calling, and information retrieval · [Required] AI Native development habits: using tools like Claude Code / Codex / Cursor as core daily development tools—exploring solutions, generating and refactoring code, completing tests, locating issues, writing documentation; also capable of strictly reviewing AI-generated code · Solid Python asynchronous programming skills: proficient in asyncio, concurrency control, timeout management, and resource isolation, with clear understanding of the boundary between 'lightweight background tasks' and 'reliable task systems' · Able to independently complete the full cycle of 'requirement → design → implementation → testing → launch → observation → iteration' · Understand the underlying logic and best practices of Web frameworks, databases, deployment and operation & maintenance, and can quickly fill gaps with the help of AI tools Preferred Qualifications · Experience with 3D modeling software, especially parametric modeling/vector 3D (training provided on entry, not required) · Practical experience in LLM application observability (e.g., LangSmith or self-built Tracing platforms) · Familiarity with modern infrastructure like Supabase, Railway, or Serverless development experience