Description
Role Summary
We're looking for an engineer to design and build the semantic and vector search layer that powers natural-language, source-grounded retrieval across a large body of government records. This role owns the retrieval pipeline end to end — from indexing through relevance tuning to answer grounding — and is central to making AI-assisted search trustworthy and auditable.
Key Responsibilities
• Design and implement hybrid retrieval pipelines combining semantic (vector) and keyword search using Azure AI Search.
• Build and tune embedding, chunking, and indexing strategies for large, heterogeneous document sets.
• Integrate Azure OpenAI to power retrieval-augmented generation (RAG) with source-grounded, citation-backed responses.
• Tune search relevance and evaluate retrieval quality against defined accuracy benchmarks.
• Collaborate with the Document Intelligence and SharePoint teams to ensure indexed content stays synchronized with source systems and metadata.
• Document architecture decisions and retrieval evaluation results for government stakeholders and auditors.
Requirements
Required Qualifications
• Production experience with Azure AI Search (or Cognitive Search), including semantic ranker and vector/hybrid search.
• Hands-on experience with embeddings, RAG architectures, and retrieval pipelines.
• Experience with Azure OpenAI or comparable LLM platforms in a production setting.
• Demonstrated work on chunking, indexing, and search relevance tuning.
• Experience building source attribution / citation-backed responses (prompt grounding).
• Strong SQL and API integration skills.
Preferred Qualifications
• Experience with retrieval evaluation frameworks (e.g., RAGAS or equivalent).
• Familiarity with AI guardrails, PII redaction, and prompt-injection defenses.
• Prior work on federal, government, or other highly regulated implementations.
• Experience with multi-agent orchestration (e.g., LangGraph) is a plus, not required.