Vector DB Engineer

Role Overview

This position is responsible for enabling natural-language querying capabilities on top of the Entity Graph by building and maintaining the vector indexing and retrieval layer powering GraphRAG. The engineer ensures that users can ask questions in plain language rather than querying the graph directly, focusing on delivering accurate, relevant, and well-evaluated responses.

Core Responsibilities

• Design and implement vector indexing strategies over graph projections and entity attributes, handling chunking, embedding selection, and index configurations.
• Develop and maintain embedding pipelines to generate, store, and refresh embeddings dynamically as entity and graph data evolve.
• Implement GraphRAG by integrating vector similarity search with graph structures and multi-hop traversal approaches to enable enriched, context-aware retrieval.
• Enable and fine-tune natural-language query scenarios, supporting techniques for translating natural language into graph queries.
• Design and conduct retrieval evaluations to measure relevance and answer quality, adjusting retrieval parameters accordingly based on agreed scenarios.
• Ensure provenance and traceability of retrieved answers by grounding them to source entities and edges.
• Optimize index size, query response time, and computational/token costs related to retrieval operations.
• Create detailed documentation covering retrieval architecture, evaluation outcomes, limitations, and supported query patterns.

Required Skills and Experience

• Expertise in vector databases and indexing technologies such as Azure AI Search, FAISS, pgvector, Pinecone, Milvus, or Fabric vector functionalities, including Approximate Nearest Neighbor (ANN) indexes and similarity metrics.
• Knowledge of embedding model selection, chunking best practices, and managing embedding refreshes to handle model drift.
• Experience with retrieval-augmented generation (RAG), specifically graph-aware retrieval techniques including context assembly, grounding, and citation.
• Strong programming skills in Python, PySpark, SQL, API development, and pipeline orchestration.
• Competence with Azure OpenAI or equivalent large language model platforms, including prompt design for retrieval and natural-language-to-query translation.
• Proficiency in designing evaluation harnesses to measure retrieval relevance, running A/B tests, and verifying grounding to minimize hallucinations.
• Familiarity with Microsoft Fabric, Azure AI services, and data lake platforms such as OneLake.
• A minimum of 5 years of engineering experience with at least 2 years dedicated to practical vector search and RAG deployment, including production-ready retrieval pipelines.

Additional Desirable Qualifications

• Hands-on experience with GraphRAG, integrating graph and vector retrieval.
• Knowledge of Microsoft Fabric NL2GQL or Data Agent capabilities.
• Understanding of graph databases and traversal methodologies.
• Experience managing large language model inference costs and latency at scale.

Primary Deliverables

• Vector indexing mechanism over graph projections.
• Embedding generation and refresh pipelines.
• GraphRAG retrieval solutions tailored for approved natural language query scenarios.
• Retrieval evaluation reports and tuning documentation.
• Documentation outlining supported querying patterns and known limitations.

Additional Context

This role complements the Senior GraphDB Engineer role closely, especially as GraphRAG requires an integration of graph and vector retrieval capabilities. There may be potential for role consolidation in future phases to optimize resources. In early project phases, this role may contribute to unstructured document processing and enriching entity attributes.