Senior Product Engineer (AI Agents)

Senior-level product engineer with 5+ years of production Python experience and strong TypeScript, React, and Node.js skills, capable of delivering AI features end to end in a web application.

They should be experienced in building and deploying LLM and agent systems, including RAG, tool use, multi-step reasoning, evaluations, APIs, data pipelines, backend services, and production monitoring.

This role suits someone familiar with owning AI products from research and model development through deployment, frontend integration, testing, CI/CD, and continuous improvement while collaborating effectively with Product, Engineering, and Data teams.

How would you design a production-grade RAG system for construction tender documents?

I would combine document parsing, structure-aware chunking, metadata filtering, hybrid keyword and vector search, and reranking. I would evaluate retrieval separately from generation and add citations, access controls, caching, and monitoring for latency, cost, and answer quality.

How would you build an LLM agent that navigates a multi-step tendering workflow?

I would model the workflow as explicit states with narrowly scoped tools, validated inputs and outputs, and clear stopping conditions. The agent should persist progress, handle retries and failures safely, and require human approval for high-impact actions such as sending bids or changing financial data.

How would you evaluate an AI feature before releasing it to customers?

I would create a representative, versioned evaluation set from real workflows and define metrics for retrieval accuracy, factuality, task completion, latency, and cost. I would combine automated evaluations with expert review, regression tests, and a limited production rollout using feature flags.

How do you reduce hallucinations in an LLM-powered application?

I would ground responses in retrieved, authoritative data and require structured outputs validated against schemas. I would also instruct the model to abstain when evidence is insufficient, expose source references, and monitor unsupported claims through automated and human evaluations.

How would you expose an AI backend service to a TypeScript and React application?

I would define a versioned API with typed request and response schemas, explicit error states, authentication, authorization, and idempotency where needed. Shared generated types or a contract-first approach would keep the Node or Python service and React client consistent.

How would you structure a Python service that orchestrates LLM calls and external tools?

I would separate domain logic, orchestration, provider adapters, tool implementations, persistence, and observability. Strong typing, dependency injection, asynchronous I/O, timeouts, retries with backoff, and contract tests would make the service maintainable and reliable.

How would you handle monitoring and alerting for production AI agents?

I would record traces for prompts, model versions, retrieved documents, tool calls, durations, token usage, and outcomes while redacting sensitive data. Alerts would cover technical failures and quality signals such as rising fallback rates, invalid tool calls, latency, cost, and task-success regressions.

How would you make LLM features cost-efficient without significantly reducing quality?

I would measure cost per successful task and use model routing, prompt reduction, caching, batching, and smaller models for classification or extraction tasks. I would validate each optimization against a fixed evaluation suite and production-quality metrics rather than relying only on token counts.

How would you safely release an AI feature in a trunk-based, daily-release environment?

I would keep changes small, protected by feature flags, and covered by unit, integration, contract, and Cypress end-to-end tests. I would deploy incrementally, monitor key product and system metrics, and maintain a fast rollback or disable path for model, prompt, and application changes.

How would you design an AWS and Kubernetes deployment for an AI backend?

I would containerize the service and deploy it through Kubernetes with resource limits, autoscaling, health checks, secrets management, and isolated environments. Terraform would define the infrastructure, while CI/CD would run tests, security checks, migrations, and controlled deployments with observability enabled from the start.

Cosuno is a fast-growing construction technology company transforming tendering and procurement through intelligent, data-driven software. As a Senior Product Engineer, you’ll build and ship LLM-powered agents end-to-end—from Python services, APIs, RAG and tool-use systems to TypeScript/React web experiences—while monitoring and continuously improving them in production. Work remotely from anywhere or from Berlin, using a stack including TypeScript, React, Node.js, Python, AWS, Kubernetes, Terraform, and Cypress. With a salary of €90K–€120K, flexible hours, a permanent contract, and meaningful ownership at the forefront of AI-driven construction innovation, this is an opportunity worth applying for.