Senior AI Engineer

Experteer Italy; Lombardia; On-site; EUR 70,000 - 100,000; Full time

Job summary

Experteer Italy is seeking a seasoned software engineer to translate business processes into reliable AI agents and systems. You will design end-to-end pipelines, select tools, and manage handoffs to ensure accuracy and trust in production models used by leading enterprises.

You will decompose processes into steps and decision points, build retrieval and grounding pipelines, and treat prompt engineering as an engineering discipline with tests and metrics.

Qualifications

• 5+ years of software development experience.
• 2+ years with production LLM-based systems.
• Experience with async work and FastAPI/Pydantic.
• Hands-on with at least one agent orchestration framework.
• Experience serving models via managed platforms or provider APIs.
• Production vector search experience and understanding of trade-offs.
• Strong ability to extract domain logic into testable artifacts.

Responsibilities

• Design and implement retrieval pipelines with chunking, embeddings, and grounding to source data.
• Develop end-to-end AI agent pipelines and tool interfaces.
• Engineer prompts, context handling, and versioning with testing and measurement.
• Build evaluation layers with datasets, automated scoring and regression suites.
• Instrument production systems for observability: traces, latency, costs, failure rates.
• Own end-to-end quality and diagnose failures across retrieval, reasoning, tooling, and data.
• Collaborate with business experts to translate domain logic into concrete artifacts.
• Take agents into production with platform teams and monitor live behavior.

Skills

LLM production systems

Async development

Agent orchestration

Tooling and function calling

Retrieval and grounding

Evaluation pipelines

Cloud-based model serving

Tools

FastAPI

Pydantic

LangGraph / LlamaIndex Workflows / Temporal

PGVector / vector databases

Job description

As part of the msg group, you will translate business processes into reliable, production-ready AI agents and systems. You design end-to-end pipelines, selection of tools, and human handoffs to ensure accuracy and trust in deployed models. You focus on turning complex domain logic into testable, measurable solutions with clear ownership of quality. This role combines engineering rigor with business impact, delivering scalable retrieval, grounding, and evaluation capabilities. A meaningful chance to shape how AI touches real work for leading enterprises.

• broad range of benefits
• diversity and cultural differences are cherished
• Decompose business processes into steps, tools, handoffs, and decision points for agent-based execution
• Design and implement retrieval pipelines with chunking, embedding models, hybrid search, and grounding to source data
• Treat prompt and context engineering as an engineering discipline with versioning, tests, and measurement
• Build the evaluation layer with datasets, automated scoring, and regression suites
• Instrument production systems for observability: traces, latency, costs, failure rates, and drift
• Own end-to-end quality and diagnose failures to their root cause across retrieval, reasoning, tooling, or data
• Balance model quality, cost, and latency, adapting choices as models evolve
• Collaborate with business experts to extract domain logic and translate it into testable artifacts
• Take agents into production with platform and infrastructure teams and monitor live behavior
• 5+ years software development experience, 2+ years with production LLM-based systems
• Experience with async work, FastAPI and Pydantic or equivalents
• Hands‑on with at least one agent orchestration framework (LangGraph, LlamaIndex Workflows, OpenAI Agents SDK, Temporal)
• Experience serving models via managed platforms (AWS Bedrock, Azure AI Foundry, Google Vertex AI) or provider APIs with streaming, structured output, prompt caching, rate‑limiting
• Production vector search experience (pgvector, Qdrant, Pinecone, Weaviate, Milvus, Azure AI Search) with understanding of trade‑offs
• Deep retrieval expertise (chunking, metadata filtering, query rewriting, hybrid search, reranking)
• Fluent in tool and function calling, structured output, and multi‑step agent design
• Hands‑on with Model Context Protocol (MCP): building servers, tool interfaces, auth and error handling
• Evidence of a solid evaluation discipline and a measurable eval harness
• Ability to work with business experts to produce workable specifications
• Judgment to select appropriate tools vs rules engines and plain queries