Business Overview

Our client is a leading US-based healthcare technology provider specializing in hub services—a sophisticated support ecosystem that streamlines patient access to specialty medications. In the high-stakes pharmaceutical sector, the company’s mission is to help navigate the administrative friction that often stands between a patient and life-critical therapy, thereby improving time to treatment initiation (TTI) by every moment saved.

The patient journey begins with an enrollment form submitted by a physician’s office. However, because every drug manufacturer and brand uses proprietary forms with unique layouts, labels, and field groupings, there is no universal document format.

Previously, parsing these documents was a grueling, highly inefficient engineering bottleneck. An entire operations team had to manually audit each unique form format, find critical data fields by eye, and manually write custom code to map those points to the client’s internal database. This approach was highly unsustainable and error-prone.

Facing a sharp seasonal spike in workload, the client approached NIX, their long-standing partner, to support their manual efforts. Recognizing the limitations of adding more headcount to a labor-intensive process, our experts instead proposed transforming this fractured, document-heavy workflow with an AI-driven enrollment form processor, pioneering a new standard for healthcare document automation.

Project Scope

We initially developed a PoC to demonstrate how manual data mapping could be automated using AI. The PoC was approved as the architectural foundation for a future smart document intelligence service.
Working alongside the client’s team, we handled the end-to-end delivery—architecting the AI logic, building the visual processing pipeline, and deploying a serverless AWS environment. The final scope included establishing a dedicated testing and observability framework to ensure the system could meet the rigorous accuracy requirements of the healthcare industry.

Challenge

During this project, our team had to navigate both technical and organizational hurdles:

Contextual mapping precision

The LLM must surpass simple guessing to accurately link visual fields to a database of hundreds of internal data points. This requires understanding a broader context of document layouts, section headings, and specialized medical fields such as “Prescriber NPI” and “No Known Allergies”.

Environmental constraints

Deploying accurate PDF rendering on AWS Lambda required a custom solution that would prevent the production of garbled or blank images that deprive the LLM of the visual context needed for mapping.

Maintaining high-quality standards

As errors can directly impact patient care and financial records, we had to achieve superior accuracy targets, proving the system’s reliability for mission-critical healthcare data.

Solution

Our team led the end-to-end development of a patient enrollment automation solution, moving rapidly from a PoC to a production-grade automation engine. The platform transforms unstructured, varied enrollment forms into standardized data maps using a highly flexible, multi-layer architecture that adapts seamlessly to any business environment.

Advanced Visual Processing Pipeline

02

To ensure intuitive, context-aware processing, the system interprets the document’s spatial layout—analyzing how text, images, checkboxes, and tables are positioned relative to one another just as a human reader would.

  • Visual logic and contextualization: The pipeline captures high-resolution digital snapshots, programmatically highlighting and indexing fields. This gives the AI full visual hierarchy to navigate complex medical forms that traditional text-scraping tools fail to read, minimizing data errors and eliminating expensive third-party licensing.
  • Dynamic prompt engineering: Our custom prompt builder coordinates business logic, user templates, and clinical data rules in real time, equipping the AI to make instantly accurate mapping decisions.

Multi-provider LLM Orchestration

To avoid vendor lock-in and ensure regulatory compliance, we implemented a provider-agnostic AI layer.

  • Universal AI interface and data sovereignty: A plug-and-play interface allows the client to hot-swap between major LLM ecosystems (OpenAI, Anthropic, Gemini, AWS Bedrock) via simple configuration changes without rewriting code. To maintain absolute data sovereignty, workflows primarily route through Claude via AWS Bedrock, keeping all protected health information (PHI) within the client’s secure, HIPAA-compliant environment.
03

Serverless Infrastructure and Cloud Automation

04

We deployed the processor as a high-performance, cloud-native microservice built for seamless integration with existing internal operator dashboards.

  • Automated and high-fidelity cloud infrastructure: Built using Terraform for Infrastructure as Code, the system features a custom cloud execution layer with embedded font libraries. This prevents image distortion and ensures the AI always receives crisp, perfectly rendered documents for error-free processing under strict security and access controls.
  • Turnkey deployment automation: A comprehensive DevOps suite enables the client to deploy, scale, or safely tear down the entire cloud infrastructure with a single command, thus lowering maintenance overhead.

Quality Assurance and Enterprise Observability

In healthcare, precision is a prerequisite. Our team moved beyond a simple prototype to create a rigorous observability ecosystem.

  • Enterprise AI observability: Integrated with Langfuse, the platform provides end-to-end visibility into every AI decision, empowering the client’s team to track performance, run quality assurance grading, and continuously optimize accuracy.
  • Cost-efficient debugging tools: Built-in debugging modes allow engineers to inspect form parsing and layout logic without running live AI queries, lowering iteration costs and speeding up manual verification cycles.
05

Outcome

As a result of this collaboration, our client received a fully functional AI form processing service that autonomously converts diverse enrollment forms into structured, actionable data. The solution is currently deployed on the company’s testing AWS infrastructure to process real-world enrollment forms and streamline the operations team’s workflow.

The project is now in an active refinement phase for final production deployment and UI integration. Following a thorough handover, the client’s internal .NET team is fully equipped with the code and documentation to scale the tool independently.

  • 90% faster processing, reducing new drug enrollment form turnaround times from several days to just a few hours.
  • 65% of the workload is automated, allowing the team to handle seasonal spikes without adding headcount.
  • Up to 95% accuracy achieved, elevating field-mapping precision from an initial 70% in the PoC to a highly reliable, production-ready standard.
Team:

Team:

4 experts ( Account Tech Lead, Tech Lead, Senior Developer, Project Manager )
Tech stack:

Tech stack:

PHP, JavaScript, Anthropic Claude, OpenAI, Google Gemini, AWS Bedrock, Langfuse, AWS Lambda, Terraform, Docker, AWS CloudWatch

REQUEST A CONSULTATION

Contact us   

Relevant Case Studies

View all case studies

AI-powered Search Solution for a Healthcare Company

Healthcare

Success Story AI-powered Search Solution for a Healthcare Company image

AI Chatbot for Personalized Healthcare

Healthcare

Success Story AI Chatbot for Personalized Healthcare image

Voice AI in Healthcare: Faster Specialty Medication Access

Healthcare

Success Story Voice AI in Healthcare: Faster Specialty Medication Access image

Clinical AI Assistant on AWS Bedrock: 40% Faster Workflows

Healthcare

Success Story Clinical AI Assistant on AWS Bedrock: 40% Faster Workflows image

Pharmaceutical Software Modernization and AI Implementation

Healthcare

Success Story Pharmaceutical Software Modernization and AI Implementation image

NLP Chatbot for Accelerating Internal Operations

Healthcare

Success Story NLP Chatbot for Accelerating Internal Operations image
01

Contact Us

Accessibility Adjustments
Adjust Background Colors
Adjust Text Colors