AI Engineer, LLM pipelines

01The role

Every call and email that enters Intract goes through a chain of LLM steps: summary, sentiment, urgency, complaint detection, translation and routing. On top of that, agents write daily digests, cluster issues into monthly reports and answer questions in the assistant.

You will own the quality, cost and reliability of that intelligence layer.

02What you will do

• Build and evolve LLM pipelines and agents with LangGraph and LangChain (TypeScript and Python)
• Run evaluations on real conversations and turn regressions into tests
• Trace every run with Langfuse and OpenTelemetry, and keep cost and latency in check
• Work on retrieval with pgvector and on topic clustering (embeddings, HDBSCAN, UMAP)
• Make every AI output traceable back to the source transcript

03What we look for

• 3+ years of software engineering, with at least one LLM-based system in production
• Strong Python or TypeScript, ideally both
• A rigorous approach to evaluation: datasets, metrics, and knowing when a prompt change made things worse
• Good understanding of structured outputs, tool calling and prompt design
• Fluent English; French or Dutch is a plus

04Nice to have

• Azure OpenAI and enterprise constraints (EU data residency, content filters, quotas)
• Speech-to-text pipelines and multilingual text
• Classical ML or NLP background

05What you will work with

• Azure OpenAI
• LangGraph
• LangChain
• Langfuse
• OpenTelemetry
• Python
• FastAPI
• pgvector