Senior Backend Engineer – Agents

This is a fully remote position, open to applicants in Germany.

📋 Description

• Create autonomous inbox agents that react to new ticket events.

• Design agent templates and establish guidelines for when agents should transfer tasks to human operators.

• Ensure compliance of agents with business processes.

• Develop harness engineering for over 100 agent tools.

• Craft system prompts and skills that enable agents to comprehend property management.

• Enhance knowledge-base learning capacities and search functionalities.

• Develop voice agents based on LiveKit.

• Build and refine LLM infrastructure.

• Choose appropriate models, caching strategies, and streaming methods to minimize latency.

• Implement resilient retry logic for provider outages.

• Manage compact context, optimize RAG search, and monitor usage across providers.

• Lower costs and assess model enhancements.

• Utilize TypeScript, Convex, Next.js, React, Tailwind, and shadcn/ui.

⛳️ Requirements

• Proficient in both English and German.

• Over 2 years of experience in developing agents for production use.

• AI-native, having delivered agent-centric features relied upon by real users.

• Possess strong viewpoints on retrieval, evaluations, tool design, and the limitations of current frontier models.

• Primarily a builder; favor releasing a fast, preliminary version over extensive planning periods.

• Stay informed about new AI models and research developments.

• Write the majority of code with the assistance of AI.

• Comfortable working independently and taking complete ownership of tasks.

• Excel particularly well under pressure.

• Quick learner who identifies as a generalist.

• Open to working with the TypeScript-based stack, as well as the Vercel AI SDK and Convex Agent Component.

🏝️ Benefits

• Fully remote work environment.

• Flexible working hours.

• Unlimited AI tokens available as needed.

• A bootstrapped and profitable company.

• Small team with genuine responsibilities.

• Swift decision-making, no unnecessary meetings, and no approval bottlenecks.