← Our Work·Internal ProductIn Active Development

Data Use Agreement Lifecycle Platform

From a single requirements document to a working, multi-tenant SaaS platform, built by a small, expert-led team using a spec-driven, AI-agent-assisted engineering pipeline. Developed internally ahead of a prospective FDA opportunity, demonstrating the same delivery approach behind our active federal engagements.

AI-Driven DevelopmentData GovernanceTransformationCloud-NativeComplianceSaaS ArchitectureDigitization

Expert-Directed

AI-Executed · Small, Senior-Led Team

9

CDK Infrastructure Stacks

12+

AWS Services Integrated

80+

API Endpoints Built

The Challenge

Data Governance Trapped in Email and Spreadsheets

Organizations that share sensitive data (research institutions, healthcare systems, government agencies) are legally required to execute Data Use Agreements (DUAs) before any data transfer. But the reality of managing those agreements was painful, manual, and fragile.

There was no system of record. No audit trail. No enforcement. Just email threads, PDF attachments, and spreadsheets that someone had to remember to update. The pain was real, and it was slowing down programs that depended on timely data access.

Data Use Agreements were negotiated over email chains with no version control, so teams lost track of what was agreed, when, and by whom

Signatures were collected as PDF attachments, offering no legal traceability and requiring manual re-filing for every amendment

Agreement status lived in spreadsheets: no centralized view, no automated reminders, no audit trail

Data access restrictions agreed in a DUA had no automated enforcement; once signed, compliance was purely manual

Cycle times stretched to weeks, delaying research programs and data sharing initiatives that depended on timely approvals

Scaling to dozens of simultaneous active agreements required proportionally more staff, which is an unsustainable model

Federal Applicability

The architecture patterns and delivery methodology used here are directly applied to active programs at FDA and ED, and completed VA past performance. The same expert-directed, AI-executed pipeline that built this platform is the foundation of our federal technology subcontracting capability.

How We Build with AI

The Multi-Agent Pipeline

One requirements document. Six specialized AI agents. A production platform. Each agent performs a discrete role: no context drift, no assumption shortcuts.

Input

requirements.md

Requirements

spec-requirements

EARS-format acceptance criteria derived from a single requirements document: complete, unambiguous, testable.

requirements.mdAcceptance criteria

Architecture

spec-design

Complete system architecture, data models, and API contracts, with every architecture decision documented before a line of code.

Acceptance criteriaArchitecture + API contracts

Task Planning

spec-tasks

Atomic, reviewable implementation tasks breaking architecture into discrete, dependency-ordered work units.

Architecture docsImplementation task list

Implementation

spec-impl

Production code generated per task: pure feature logic across backend, frontend, database, and infrastructure.

Task listProduction code

Security Audit

spec-review

Autonomous security review: RBAC, RLS, OWASP Top 10, injection risks, and access control, applied to every implementation task.

Production codeSecurity-hardened code

Test Generation

spec-test

Independent test suites generated by a separate agent: unit tests, Playwright E2E tests, and a full backend test suite.

Security-reviewed codeProduction platform + test suites

Output

Production Platform

The Solution

End-to-End Data Agreement Governance

A secure, multi-tenant SaaS platform that manages the full lifecycle of Data Use Agreements, from drafting through enforcement.

Full DUA Lifecycle Management

From creation through negotiation, signing, amendments, and expiration: the entire agreement lifecycle in one place. Auto-generated DUA numbers, HHS DUA Policy fields pre-built, status tracked at every step.

Multi-Party Clause Negotiation

Both parties discuss and negotiate individual terms inline with threaded comments, with no email required. Every round of negotiation creates a new version with a complete edit history.

Digital Signatures with Legal Traceability

Canvas-based drawn signatures with exact timestamps and immutable signing history. Not a checkbox: a real signature, captured with full audit context.

Machine-Readable Rule Export

On signing, agreements are automatically exported as structured JSON policies, enabling external data platforms to programmatically enforce DUA restrictions without manual intervention.

Immutable Audit Trail

Every action (DUA creation, status changes, term edits, signatures, comments) is logged at the database level. Records cannot be altered or deleted. Compliance-ready from day one.

Multi-Tenant Architecture

Schema-level tenant isolation, not just row-level filtering. Multiple organizations share one platform with zero cross-tenant data leakage and independently configurable workflows.

Also Delivered

Study OrdersDUA TemplatesEmail NotificationsReminder CronsTenant BrandingWebhook EventsDatabricks ExportPlatform Admin

AI Value-Add

What Our AI Approach Added Beyond Code

The AI didn't just write code. It brought architectural rigor, security discipline, compliance intelligence, and operational quality that traditional development rarely achieves consistently.

Architecture-First Thinking

Architecture decisions documented before a single line of code, ensuring system integrity by design.

Security by Default

OWASP Top 10 reviewed on every implementation task. Security isn't a phase; it's built into every step.

Compliance Intelligence

Full HHS DUA Policy (HHS-OCIO-CDO-2023-01-001) compliance built in, not bolted on after delivery.

Multi-Tenant Isolation

Schema-per-tenant, RLS on every table. Isolation at the database level, not just the application layer.

Async Event Architecture

Multiple notification event types via SES + SNS async pipeline: scalable, reliable, loosely coupled.

Living Documentation

5 role-specific user guides auto-generated: Admin, Data Provider, Data Recipient, Reviewer, Auditor.

Zero-Downtime Migrations

23 Alembic migrations: every schema change forward-only, tested, and zero data loss.

Full Infrastructure as Code

9 CDK stacks, 12 AWS services, reproducible environments: dev, staging, and production from one codebase.

Self-Improving Quality Loop

3-agent review cycle on every change: implement → security review → independent test generation.

Expert-Validated Output

Every AI-generated artifact, including architecture decisions, implementation tasks, security reviews, and test suites, was reviewed and approved by the senior architect before integration. AI executes; the expert decides.

The Impact

Architecture Discipline, Built with a Small, AI-Assisted Team.

Small Team

Senior-led delivery

AI-agent-assisted pipeline, not a large multi-role team

9

CDK infrastructure stacks

Reproducible dev, staging, and production environments

80+

API endpoints built

Across 12 AWS services

Multi-Suite

Automated test coverage

Unit, integration, and end-to-end suites covering the core platform

Low-Cost

Infrastructure cost

Pay-per-request · scales with usage · multi-tenant

Multi-Layer

Codebase

Backend, frontend, and infrastructure-as-code

20

Database tables

Row-Level Security enforced on every one

Documented

Architecture decisions logged

Before a single line of code was written

By the Numbers

A Different Kind of Development

The direct comparison: what this engagement delivered vs. what traditional development would have required.

Traditional Approach

This Platform

Internal Build, In Development
Team
Backend, frontend, DevOps, QA, PM, architect, security: larger multi-role team
Expert-Directed · AI-Executed pipeline, fully documented and transferable
Delivery Approach
Sequential handoffs across specialized roles
Spec-driven, AI-agent-assisted pipeline with expert review at every step
Maintenance
Requires ongoing dedicated headcount across roles
Maintained by the same senior architect and AI-assisted engineering pipeline
Test Coverage
Variable
Unit, integration, and E2E suites
Security
Added after build
Built in from day one
Documentation
Often incomplete
Auto-generated, role-specific

Technologies Used

The Stack Behind the Platform

Frontend

  • Next.js 15 (App Router)
  • React 19
  • Tailwind CSS 4
  • shadcn/ui
  • TanStack Table
  • React Hook Form
  • TipTap (Rich Text)

Backend

  • FastAPI (Python 3.12)
  • SQLAlchemy 2.0 (async)
  • PostgreSQL 16
  • Pydantic 2.7
  • Alembic (23 migrations)
  • pytest / pytest-asyncio

AWS Cloud (12 Services)

  • Lambda (API + Email)
  • RDS PostgreSQL 16
  • AWS Cognito (Auth)
  • S3 (Storage)
  • CloudFront (CDN)
  • SES (Email)
  • SNS / SQS
  • Amplify
  • ECR (Docker)
  • EventBridge
  • Secrets Manager
  • CloudWatch

Security & Compliance

  • JWT RS256 / JWKS verification
  • Row-Level Security (RLS)
  • Schema-per-tenant isolation
  • HHS DUA Policy fields
  • Immutable audit logging
  • AWS Secrets Manager
  • 3-tier VPC network

DevOps & Testing

  • GitHub Actions + OIDC
  • Infrastructure-as-Code (CDK)
  • Vitest (unit tests)
  • Playwright (E2E tests)
  • Multi-environment (staging / prod)
  • Automated deployment pipeline

This engagement is managed under the direction of Shweta Nagpal, Founder & President, elfOvations LLC.

Have a Similar Requirement?

This platform demonstrates how an expert-directed, AI-executed pipeline pairs senior architecture with a small, AI-agent-assisted delivery team. Tell us your requirement and we will show you what is possible.