About the role
You will set technical direction, make the hard model and system calls, and be accountable for quality when the system is live in front of customers. This is a high-ownership role for someone who has shipped applied ML and LLM systems into production, watched them break on messy real-world data, and rebuilt them to hold.
Responsibilities:
- Own the core ML and AI systems of the product and the technical direction behind them.
- Build retrieval, ranking, and LLM systems over large volumes of real operational data.
- Design LLM workflows and agentic flows that produce structured, reliable outputs and operate safely against production data.
- Build the evaluation harness and guardrails: measure grounding, consistency, and hallucination, price the cost of a wrong answer, and catch bad outputs before a user sees them.
- Own production health: latency, inference cost, output consistency, observability, and the tuning cadence that keeps quality from degrading.
- Raise the engineering bar for the team. Engineers will look to you for how to decide, not just what to do.
Requirements:
- 4+ years building and shipping applied ML or LLM systems that ran in production.
- Strong Python and solid ML fundamentals.
- You understand why a system behaves the way it does, not just which function to call.
- End-to-end ownership of an ML or LLM pipeline: data, modelling or orchestration, evaluation, deployment, and the follow-through of improving quality in production.
- Hands-on production experience with LLM systems: RAG, agents, structured outputs, and disciplined prompt engineering backed by evaluation.
- Depth in retrieval: embeddings, vector search, and semantic retrieval over messy real-world data.
- Experience with ranking, recommendation, or personalisation systems, and the judgment to reason about relevance and ordering.
- Backend competence to ship the whole feature: APIs, SQL and relational databases, cloud deployment, logging, and observability.
You'll stand out if you have:
- Evaluation or guardrail systems you have built for LLM outputs, especially around hallucination, grounding, and consistency.
- Breadth across the current AI stack: orchestration frameworks, vector databases, in-memory stores, streaming event processing, and LLM observability tooling.
- Memory, context management, or follow-up resolution for a conversational system at scale.
- Experience mentoring engineers and raising a team's standards.