Role Purpose
Deliver the natural-language query capability over the Entity Graph. This role implements the vector indexing and retrieval layer that powers GraphRAG — enabling users to ask questions in plain language instead of writing graph queries — and is accountable for the accuracy, relevance and evaluation of those responses.
Key Responsibilities
- Vector index design & build — design and implement the vector indexing strategy over graph projections and entity attributes, including chunking, embedding selection and index configuration.
- Embedding pipeline — build pipelines to generate, store and refresh embeddings as entity and graph data changes.
- GraphRAG implementation — combine vector similarity search with graph structure and multi-hop traversal to produce grounded, context-rich retrieval.
- Natural-language query enablement — implement and tune NLQ scenarios agreed with WK; support natural-language-to-graph-query translation approaches.
- Retrieval evaluation & tuning — define and run evaluation harnesses measuring retrieval relevance and answer quality; tune retrieval parameters against agreed scenarios.
- Grounding & traceability — ensure retrieved answers are attributable to source entities and edges, preserving provenance.
- Performance & cost management — optimise index size, query latency and compute/token cost of retrieval operations.
- Documentation — document retrieval architecture, evaluation results, known limitations and supported query patterns.
Must-Have Qualifications
- 5+ years engineering experience with 2+ years hands-on vector search / RAG implementation
- Demonstrable production experience building a retrieval pipeline (not prototype-only)
- Solid Python skills and familiarity with embedding models
- Experience evaluating and tuning retrieval quality with defined metrics
- Understanding of how to ground responses and preserve source traceability