About the Role:

This role sits at the intersection of business, analytics and applied AI. You'll work directly with business teams across sales, supply chain, finance, commercial and marketing diagnosing problems, designing systems and shipping solutions that real teams depend on. You own problems end to end: from understanding the business rules to writing the code, deploying it, monitoring it and documenting it so others can build on it.

What You'll Do:

- Build end-to-end automations that replace manual workflows from trigger to delivery, with approval steps where needed.

- Build LLM agents and RAG systems over structured and unstructured data, with grounded and cited outputs.

- Write production SQL and Python to extract, transform and reconcile data every output validated before it reaches a stakeholder.

- Deploy and maintain services on GCP with logging and alerting in place.

- Translate business rules into documented code that non-engineers can verify and sign off.

What we're looking for:

- Python: Production-grade Python: scheduled services, background workers, data pipelines and API integrations. Config-driven, tested and reliable. Not notebooks.

- SQL and BigQuery: Strong SQL for transformations, modelling and reconciliation at scale. Declares grain before joining, validates outputs against source.

- LLM engineering: Has shipped LLM applications to real users with tool calling and structured outputs. Treats prompts as code - versioned and tested against labelled samples. Accuracy measured before and after every change, not assumed.

- RAG and knowledge systems: Has built RAG systems in production: ingestion pipelines, embeddings, vector databases and hybrid search. Evaluates retrieval quality on a held-out set.

- Cloud and deployment: Has owned and operated services on GCP end to end. Responsible for production uptime, not just deployment.

- Data validation and testing: Writes automated checks for every pipeline: row counts, source reconciliation, freshness and edge cases. A system is not done until it can detect its own failures.

Good to have:

- Agent frameworks (LangGraph, CrewAI) or MCP servers.

- Event-driven and webhook integration patterns at production scale.

- Semantic layer or metrics catalog (dbt, LookML, Unity Catalog).

- Tech product or FMCG / consumer industry background.

- System integration - connecting AI outputs into existing tools and business systems (ERP, CRM, BI dashboards, internal platforms).

- Can read a data scientist's notebook, understand the logic and convert it into a production service.