ML Engineer: LLM Systems for Investment AI (London Office)

Descrial; Greater London; Sur place; GBP 90 000 - 150 000; Plein temps

Résumé du poste

Descrial is hiring a Machine Learning Engineer to design, build, and evaluate LLM-powered systems at the core of W-OS. You will work with LangGraph and LangChain on Python/FastAPI backends, focusing on multi-step agentic workflows, retrieval over confidential corpora, and production observability.

You will join a small, senior team with mentorship to shape architecture, cost-aware context handling, and robust model routing.

Qualifications

• 5+ years of software engineering with strong Python.
• 2+ years building LLM/ML systems in production.
• Production RAG experience with embeddings and vector databases.
• Graph-backed retrieval experience or fundamentals to pick up Neo4j/Cypher.
• Evidence of evaluation-led development.
• Experience using AI coding tools daily.

Responsabilités

• Design and implement multi-step agent workflows with LangGraph/LangChain.
• Build structured-output and prompt systems (Pydantic-validated).
• Integrate model providers (Anthropic, OpenAI, Bedrock) with routing/fallback.
• Manage context, token costs, and model selection for production.
• Develop MCP server services for safe access to internal tools.
• Own end-to-end RAG pipelines and knowledge-graph retrieval.

Connaissances

Python programming

LLM/ML in production

Agent/orchestration experience

Evaluation-driven development

AI tooling usage (Cursor/Claude Code)

Formation

Bachelor's degree in engineering / CS or equivalent

Outils

LangGraph

LangChain

Neo4j/Cypher

Qdrant/Weaviate/pgvector

Description du poste

Descrial is hiring a Machine Learning Engineer to design, build, and evaluate LLM-powered systems at the core of W-OS. You will work with LangGraph and LangChain on Python/FastAPI backends, focusing on multi-step agentic workflows, retrieval over confidential corpora, and production observability.

You will join a small, senior team with mentorship to shape architecture, cost-aware context handling, and robust model routing.