Gene Target Prioritization for Obesity Therapy Research
Healthcare
Target Synergy and Clinical Insight Discovery platforms for immunology research.
A healthcare revenue cycle management company needed a better way to manage accounts receivable operations across post-acute, skilled nursing, and long-term care facilities. The goal was to build a scalable, multi-tenant web platform that would bring AR balances, cases, claims, tasks, automation, dashboards, reporting, and administration into one environment. Today, the platform is live across almost 300 facilities. AI plays a central role in how the platform is developed. The seven-person engineering team uses AI agents and AI-assisted tools throughout the software development lifecycle—from requirements intake and codebase research to implementation, testing, troubleshooting, review, documentation, and pull-request preparation.
The client is a healthcare revenue cycle management company serving post-acute, skilled nursing, and long-term care providers.
- Healthcare
- Ongoing
- 7 developers: 5 backend and 2 frontend
Product
Multi-tenant accounts receivable management platform
Production scale
Almost 300 facilities
Users
Billers, supervisors, managers, trackers, and administrators
Architecture
Multi-tenant, event-driven platform with Angular micro frontends
AI approach
AI tools are used throughout the engineering lifecycle
Project Scope
The platform brings core AR workflows into one environment. Its implemented functionality includes AR aging and balances, cases, task and claim-status management, dashboards, reports, user and facility administration, automation rules, exports, notifications, and integrations with healthcare systems.
Tech Stack
The platform combines a .NET backend, Angular micro frontends, Azure infrastructure, and Microsoft data and reporting technologies.
Backend
- .NET Aspire
- C#
- .NET 10
- ASP.NET Core
- FastEndpoints
- Entity Framework Core
- Dapper
- FluentValidation
- Temporal
- Azure Functions
Frontend
- Angular 21
- TypeScript
- Angular Material/CDK
- native-federation micro frontends
- NgRx Signals
- RxJS
- Tailwind CSS
- Storybook
Data and reporting
- SQL Server
- Microsoft Fabric
- Power BI semantic models
- DAX
- Redis
- Blob Storage
- search
Cloud and infrastructure
- Microsoft Azure
- Azure Service Bus
- Azure Storage
- Azure Functions
- .NET Aspire orchestration
- Infrastructure as Code
Security
- Microsoft Entra ID/MSAL
- OpenFGA
- organization scoping
- audit logging
- soft deletion
Integrations
- PointClickCare
- NCS
- healthcare billing data sources
- Stedi healthcare APIs
- Microsoft Fabric/Power BI
- SendGrid
Testing
- xUnit
- Angular tests
- Storybook
- Playwright
- ESLint
- code coverage
The development workflow also uses the GitHub Copilot extension in VS Code, GitHub Copilot in the Copilot app and GitHub, project-specific agent skills and prompts, child agents, AI-assisted pull-request review, and Claude, GPT, and Grok models.
The Client
The client supports healthcare providers with revenue cycle and accounts receivable operations. Its teams work with unpaid balances, claims, payer information, collection tasks, and other follow-up activities across multiple facilities and payer types. The company needed one scalable platform to replace fragmented AR workflows and evolve processes previously handled through a legacy system. The platform serves several user groups.
- AR billers follow up on unpaid balances and collection tasks.
- HMO, Medicare, and Medicaid billers handle accounts associated with specific insurance or government payer types.
- Supervisors and managers assign work and monitor team performance.
- Trackers monitor case progress and required follow-ups.
- Organization administrators manage users, roles, facilities, and system settings.
Business Requirements
The client needed a single AR workspace that could bring together operational data and translate business rules into actionable work.The core requirements were to:
- provide one workspace for AR operations;
- turn business rules into tasks;
- automate repetitive work;
- integrate with billing systems;
- protect tenant data;
- provide operational and financial reporting.
The Problem
AR teams previously relied on several systems and manual processes to understand unpaid balances and decide what to address next.Information about residents, payers, claims, and tasks was not available in one place.This fragmented workflow slowed day-to-day operations, increased the risk of mistakes, and made it harder to track progress consistently.The client needed a platform that could consolidate this information while transforming complex AR rules into manageable, trackable workflows.
The Challenges
- One of the biggest technical challenges was turning client-specific AR and billing rules into reliable software workflows.
- The application also has to synchronize legacy data and information from several healthcare systems. Some workflows may run for hours and process large volumes of records.
- Temporal was selected for durable workflow execution, while queues and background services handle long-running imports, exports, and automation without blocking the application.
- Scale is another challenge. New facilities are onboarded regularly, significantly increasing the data the platform must process while maintaining reliable performance.
- Security requires controls at multiple levels. Organization scoping prevents one tenant from accessing another tenant's data, while OpenFGA controls access to specific facilities and features.
- Because the platform handles sensitive healthcare and financial information, authorization, audit logging, and PHI/PII protection are also part of the architecture.
- On the frontend, micro frontends let business areas develop and release independently. This supports scalability and parallel delivery but requires stable API contracts, shared components, and coordinated dependency management.
- The engineering team also has to keep frontend modules, backend services, infrastructure, tests, and releases aligned across the project.
Problems Revealed
As the product grew, the team identified opportunities to improve access control, performance, data synchronization, and delivery processes.These findings led to targeted technical improvements, stronger testing, and clearer review processes.
The Solution
The platform uses an Angular native-federation shell with domain micro frontends.The .NET backend separates APIs, Temporal workflows and workers, Azure Functions, reporting, authorization, auditing, and database deployment.SQL Server stores operational data. Azure Service Bus, Blob Storage, Redis, and Temporal support background processing, while Microsoft Fabric and Power BI support reporting.Five backend and two frontend developers work through a structured development process that includes Jira-driven research, planning, human approval, implementation, focused testing, independent review, and pull-request approval.
AI Across the Engineering Lifecycle
AI is used throughout the development process, not just for autocomplete or occasional coding assistance.The team initially adopted tools such as the GitHub Copilot app and the GitHub Copilot extension for VS Code and gradually expanded their use as AI models and development tooling improved.Once a recurring process stabilizes, the team can formalize it as a reusable, project-specific skill so AI agents can follow established conventions consistently.
AI analyzes business requests, drafts and creates Jira items, identifies missing information, and turns approved requirements into implementation plans.
Agents explore the codebase, trace dependencies and data flows, compare implementation options, and support engineers as they evaluate architectural decisions.
AI generates or updates production code, tests, database changes, infrastructure configuration, and documentation. Writing implementations manually, line by line, is no longer the team's default workflow.
AI creates and runs tests, investigates failures, reviews pull requests, identifies defects, and applies approved fixes.
Agents analyze logs, errors, failed workflows, CI results, and behavior across services to identify root causes and propose or implement corrections.
AI profiles SQL and DAX queries, reviews execution behavior, identifies bottlenecks, and supports optimization of workflows, APIs, and data processing.
AI analyzes telemetry from application environments to identify unusual behavior, recurring failures, capacity concerns, and broader ecosystem-health risks.
AI prepares pull requests, release notes, and technical documentation. It can also analyze pull-request comments, implement approved changes, reply with the resolution, and follow project-specific skills and conventions.
Engineers and other project specialists remain responsible for supplying business context, defining requirements, making architectural decisions, validating results, and approving production changes.
How the AI-First Workflow Differs From Traditional Development
In a traditional development process, engineers manually produce most code, tests, tickets, investigations, and documentation.In this project, AI produces or accelerates many of these outputs across the development lifecycle. The team directs the work, supplies domain knowledge, evaluates trade-offs, and validates quality, security, and business behavior.The result is not autonomous software development. Human approval remains required for production code and merges. The workflow does, however, support a high and steadily increasing delivery volume.
Achievements
The platform is now live across almost 300 facilities. AI accelerates research, implementation, review, debugging, testing, and documentation across the engineering workflow. AI-assisted pull-request review has also helped identify difficult defects.The approach supports faster development and better parallel delivery between frontend and backend work. Reusable project skills and AI-assisted review also help improve consistency and knowledge sharing across the engineering process.Development throughput continues to grow as the team gradually adopts AI across the engineering lifecycle. Before AI adoption, the team averaged approximately 4–5 PRs per business day, compared with 8.7 PRs per business day in August 2026.In August 2026:
- 182 pull requests were created;
- 173 pull requests were merged;
- the repository averaged 8.7 PRs per business day;
- this equals an estimated 1.2 PRs per developer per business day for the seven-person engineering team.
- A straightforward background job for transaction processing with no architectural changes went from approximately 1–1.5 days before AI adoption to 2–4 hours.
- An error-handling architecture change for background jobs, including tests, went from approximately 4–5 days to around 2 days.
- A CRUD endpoint with tests went from approximately one day to 1.5–2 hours.
The Impact
The impact can also be seen in the time required for specific development tasks:
- A straightforward background job for transaction processing with no architectural changes went from approximately 1–1.5 days before AI adoption to 2–4 hours.
- An error-handling architecture change for background jobs, including tests, went from approximately 4–5 days to around 2 days.
- A CRUD endpoint with tests went from approximately one day to 1.5–2 hours.
Future Plans
This case study does not include the public product roadmap because future capabilities have not been confirmed for disclosure. Within the engineering process, the team continues refining its AI workflow, reusable skills, review practices, governance, and measurement. No additional product AI capabilities are included here until they are implemented and approved for public disclosure.
Ready to see how enterprise-grade security can accelerate your AI initiatives?
Schedule a Technical Security Briefing