What you will do

• Design and deliver production agents and multi-step LLM workflows in TypeScript and Node.js.
• Define reliable tool contracts, structured outputs, retries, timeouts, approval boundaries, and human escalation.
• Build retrieval pipelines covering ingestion, chunking, embeddings, indexing, retrieval, reranking, grounding, and citations.
• Develop backend APIs and services with NestJS and PostgreSQL.
• Create evaluation datasets and automated checks for correctness, tool use, retrieval quality, latency, and cost.
• Instrument traces, logs, metrics, token usage, and model costs.
• Investigate production failures and improve prompts, retrieval, tools, and orchestration using evidence.
• Integrate model providers behind stable application interfaces and support provider or model changes.
• Participate in client discovery, explain trade-offs, estimate work, and deliver in reviewable increments.
• Review human- and AI-generated code to the same engineering standard.

Required experience

• 5+ years building and operating production backend or full-stack software.
• Strong TypeScript and Node.js, including API design, asynchronous programming, testing, and production debugging.
• At least one substantial LLM, agent, tool-calling, or production RAG system delivered beyond the prototype stage.
• Practical experience with tool/function calling and either agent orchestration or retrieval.
• Strong PostgreSQL, SQL, and data-modelling skills.
• Ability to explain real failure modes including hallucination, poor grounding, brittle tools, loops, latency, cost, and security.
• Production ownership including CI/CD, monitoring, incident investigation, and incremental delivery.
• Strong written and spoken English for client and team communication.

Useful, not required

• NestJS.
• LangGraph, Vercel AI SDK, Mastra, OpenAI Agents SDK, or custom orchestration.
• LangSmith, Langfuse, OpenTelemetry, or equivalent evaluation and tracing tools.
• Qdrant, pgvector, Pinecone, Weaviate, or another vector store.
• MCP client or server integrations.
• Python for experiments and data-processing tasks.
• Previous consultancy, outsourcing, or direct client-facing delivery.

What we offer

• Full-remote work with a flexible schedule outside agreed client and team overlap.
• Full-time independent-contractor engagement.
• 14 paid vacation days and 7 paid sick days per year.
• Official public holidays in your country of residence in addition to the paid days above.
• Work from your own laptop; EthBerry provides the project accounts, model access, and paid software licences required for the engagement.
• Compensation agreed individually based on relevant experience.
• Direct participation in architecture and technical decisions on client projects.
• The contract does not include health insurance, company hardware, or office-based benefits.

Hiring process

1. Application review.
2. HR screening — 30 minutes.
3. Technical interview — 75 minutes. We discuss one production system, its architecture, failures, evaluation strategy, retrieval quality, and operational trade-offs.
4. Unpaid take-home assignment — strictly capped at 4–6 hours. Build a small TypeScript agent with two tools, retrieval over supplied fixtures, one malformed tool response, one no-answer case, a compact evaluation set, and structured logs. No deployment, frontend, personal API spending, or production-ready polish is required.
5. Assignment review — 75 minutes. You explain decisions and respond to one changed requirement.
6. Final cultural-fit interview — 30 minutes.

AI coding tools are allowed. Their use must be disclosed briefly, and you must be able to explain and defend every submitted decision. The assignment is not used in production.