Senior Azure Data Architect

Share this job as a link in your status update to Linked In.

Job Title

Job Type

Full-time

Category

Information Technology

Location

FAIRFAX - , VA 22030 US (Primary)

Education

Bachelor's Degree

Travel

Job Description

A federal agency already runs a production AI Assistant on Azure Government. You will be the technical authority who extends it to the case files, case records and legal documents of three more of its offices, so that an attorney or investigator gets a cited answer in seconds and never sees a page their role is not cleared to see. Full time, W-2, starting 19 October 2026.

Our customer is a federal agency that enforces federal law. Its attorneys and investigators decide real cases from thousands of pages of case files, exhibits and correspondence. The agency has built the platform to search that record with AI. What it needs now is one senior person who can bring new offices onto it without weakening the access model that protects the people named in those files.

This is a named Key Personnel position. You are the architect the agency's data and CIO staff will work with directly on design, and your resume is presented to the Contracting Officer for approval before you start. It is a builder's seat on a team of three: you, a Senior BI and Analytics Engineer, and a Program Manager who handles the customer, the schedule and the security writing so that your hours go to engineering.

You inherit a working stack and you build inside it. The agency operates the platform; you own what SPN builds on it.

• Azure Government, with separate dev and test subscriptions and production behind an agency go decision
• A governed PostgreSQL read replica fed by change data capture from the agency's case system
• A dbt semantic layer and a data catalog
• Approved Azure OpenAI services, a platform vector store, OCR and embedding services
• Apache Superset for dashboards and AI-assisted views
• A production AI Assistant with retrieval guardrails already in place
• The agency's Git repositories and CI/CD runners, where all code lives from the first commit
• Python: pandas, SQL Alchemy, psycopg, the Azure OpenAI SDK and test frameworks

• Own the technical design of every product and record each design decision in the agency repository
• Register each office's case tables and document sets as sources, with a freshness expectation for each
• Check every document for a native text layer and send only image pages to OCR
• Hold pages below the agreed OCR confidence in a review queue instead of indexing them
• Split text on headings, numbered paragraphs, exhibits and pages, and tag every chunk with its document, page, case, owning office and role scope
• Keep the index current from the change stream, removing deleted, withdrawn and redacted material explicitly
• Configure hybrid lexical and vector retrieval with rank fusion, suppress any sentence without a supporting source, and refuse when the evidence is weak
• Compile each user's identity claims into a mandatory filter on every vector query, and write PostgreSQL row-level security and field masking by role
• Write an audit record for every retrieval and export
• Measure retrieval recall, citation precision, groundedness, refusal correctness and access correctness against the agency's test set at every acceptance test
• Build the CI/CD path that promotes dbt, Python and Superset assets from dev to test to production, with a probe that checks upstream schema and credentials before every scheduled run
• Supply the configuration evidence and test results behind the NIST SP 800-53 control statements, and fix penetration-test findings on what we build

• Days 1 to 30: suitability forms filed on day one; source inventory and design for the first…