The healthtech firm wanted to expedite clinical trial protocol reviews by building an AI-powered application. Our experience with AWS healthcare services and regulatory compliance frameworks made us the right partner for this mission-critical implementation.

1. Researching existing workflows

We began by observing how analysts managed their routine operations. The research revealed that manual data lookups, repetitive text searches, and inconsistent protocol formatting slowed every step of the protocol review process. 

These bottlenecks helped us understand where automation can have the greatest impact. Everyone agreed that the new platform must be able to accept multiple document types, adapt to regional data formats, and deliver actionable insights while preserving established workflo 

2. Bringing intelligence to protocol reviews

Next, we worked on an NLP core built on Amazon SageMaker. The team analyzed millions of historical trial protocols to structure key details like endpoints, eligibility, and dosing. A pre-trained BioClinicalBERT model, developed using MIMIC-III clinical notes and fine-tuned with internal datasets, added domain-specific accuracy. 

This progression turned a static review process into one capable of automatically detecting gaps, conflicts, or unusual deviations that could influence trial outcomes.

3. Scaling insights in real time )

The intelligence layer required both speed and scale to be effective. We deployed inference endpoints on AWS Inferentia to deliver low-latency outputs, giving analysts actionable insights within seconds of uploading a protocol. In parallel, Amazon EC2 UltraClusters enabled large-scale, distributed processing of millions of protocol pages and historical datasets. 

This high-performance environment reduced tasks that once took weeks to just hours, while advanced cross-document comparisons created a more comprehensive and consistent review process across trials.

4. Building a compliant, future-proof platform

The team embedded GxP compliance into the solution from the start, aligning workflows, audit trails, and data handling with stringent regulatory standards. We used AWS Trainium to provide cost-efficient training for fine-tuning NLP models, enabling faster experimentation without inflating infrastructure costs. 

The team also enabled AI-driven coding suggestions with Amazon CodeWhisperer. This helped developers rapidly implement data pipelines, interface components, and integration scripts.

Together, these capabilities created a secure, compliant platform that was ready to integrate with real-world data sources.

5. Driving stickiness with GitHub Copilot

Modus Create's Training and Enablement team partnered with key stakeholders to design hands-on sessions aligned with both business goals and budget. Drawing on our experience scaling enablement across large teams, we delivered targeted 101 (beginner) and 201 (intermediate) workshops. We then followed up with an ROI session to help teams translate their performance, usage, and priorities into actionable metrics.