About the Organization
A federal consulting organization conducts rigorous year-end performance reviews for its workforce. For each employee, HR staff must synthesize thousands of lines of data — manager comments, self-assessments, performance ratings, goals achieved, and development areas — stored across multiple tables in Microsoft SQL Server.
Operating under strict compliance and data sovereignty requirements, the organization needed a way to apply modern AI to this HR workflow without sensitive employee data ever leaving on-premises infrastructure — and with every data access governed and auditable.
The Challenge
The organization faced a tension between operational efficiency and compliance:
Manual synthesis at scale: HR staff had to manually review and summarize thousands of lines of data per employee — manager comments, self-assessments, ratings, goals, and development areas — pulling professionals away from the review conversations that actually matter.
Data sovereignty constraints: Sensitive employee data could not leave on-premises infrastructure, ruling out any cloud-based AI service.
Deterministic access only: Any AI adoption required that the model could only access pre-approved, pre-filtered data — not generate arbitrary SQL queries against the HR database.
Auditability requirements: Every data access had to be logged and auditable, fitting within existing security and compliance frameworks without introducing new blind spots.
The Solution
DreamFactory provided the governed API layer that made secure AI-to-database communication possible — without writing custom backend infrastructure.
Deterministic data access via stored procedures: Rather than allowing the AI to generate SQL, all data access was encapsulated in a SQL Server stored procedure (review.GetEmployeeAppraisalData) that returns only the fields necessary for summarization — no salary data, no Social Security numbers, no draft reviews. DreamFactory auto-generated a secure REST endpoint with authentication, role-based access control, and full audit logging.
On-premises AI via HTTP connector: A local large language model running on an NVIDIA DGX Spark appliance was exposed through DreamFactory’s HTTP service connector as a managed API (/api/v2/AppraisalAi/chat/completions) — applying the same governance, authentication, and logging as any other DreamFactory-managed service.
Orchestration with a Python scripted API: A DreamFactory scripted service ties the workflow together in a single API call — accepting an employee identifier, invoking the stored procedure through DreamFactory’s governed API, constructing the AI prompt, calling the on-premises model, and returning the generated performance summary. The entire pipeline runs within the organization’s infrastructure. No data leaves the network. Every step is logged.
The Results
HR workflow transformed: What previously required HR staff to spend significant time manually reviewing thousands of lines of data per employee now happens in seconds through a single API call — freeing professionals to focus on conducting meaningful review conversations instead of synthesizing raw data.
Compliance preserved by design: The stored-procedure pattern ensures the AI model can only access pre-approved data through a defined security contract, satisfying federal requirements for auditable, governed data access.
True data sovereignty: Because the AI model runs on-premises on the organization’s own NVIDIA DGX Spark hardware, sensitive employee information never traverses an external network.
First-of-its-kind AI orchestration: This implementation represents one of the first production deployments of DreamFactory orchestrating an end-to-end AI workflow — from governed database access through on-premises AI processing — demonstrating that enterprises can adopt AI without sacrificing security, compliance, or data sovereignty.
