HIPAA-Safe. Patient-Smart
A healthcare digital services company set out to revolutionize patient engagement — delivering targeted medical ads and educational content based on individual health history, while staying entirely outside HIPAA scope.
The Challenge
Two hard constraints defined this project: compliance and speed.
Patient data had to be anonymized in a way that satisfied regulators without losing the ability to make personalized targeting decisions.
At the same time, the system had to process requests and return results in under 300 milliseconds — at scale, in real time.
The Solution
We built a two-track architecture — a real-time ad serving pipeline on AWS, and a campaign management application running in Docker.
- Lambda functions handle inbound requests, fan out to internal decisioning services, and return results within the latency requirement
- Third-party tokenization anonymizes patient identifiers while preserving a consistent key for targeting — keeping PHI out of the pipeline
- SNS/SQS manage async tracking of impressions, clicks, and content views
- Snowflake captures and surfaces engagement analytics
The Results
The platform had a successful launch, enabling healthcare providers and life sciences companies to connect with patients before, during, and after care visits through a 1:1 engagement experience.
"Our partnership enabled our business to function with the agility and innovation of a pre-product startup for nine crucial months — instrumental in launching our groundbreaking 1:1 patient engagement platform."
— Senior Engineering Manager
AI That Pays for Itself
A warranty and protection solutions provider was losing ground to market pressure and operational inefficiency. With no prior AI experience, they partnered with Pinnacle to modernize their claims process — incrementally, and without disrupting compliance.
The Challenge
Warranty adjusters were spending 25–100 minutes of manual research per claim — looking up parts, calling repair shops, and verifying vendor approvals by hand.
Across thousands of monthly claims and 15–20 adjusters, this added up to hundreds of hours of lost productivity every month, with significant cost savings being left on the table.
The Solution
We designed and built an AI-powered workflow integrated directly into the client's existing claims platform.
Perplexity AI searches for alternative OEM and aftermarket parts across multiple sources simultaneously
GPT-4 (private instance) validates part fitment and compatibility against manufacturer specs
Automated blacklisting and vendor filtering keeps results compliant with approved supplier lists
Audit trails and compliance checks are baked in — not bolted on
Azure Functions handle serverless, scalable batch processing
The Results
- Processing time dropped from 25–100 minutes to ~8 seconds
- 4x improvement in adjuster productivity
- Projected annual savings of $1–3 million
- Adjusters freed from repetitive lookups to focus on higher-value decisions
- Foundation established for continued AI expansion across the business
"This AI solution represents exactly the kind of operational efficiency we need. Pinnacle's incremental approach gave us confidence to embrace AI transformation while maintaining our compliance standards."
— VP of Operations, Claims & Compliance
Faster Releases. Fewer Fires.
A healthcare services company was struggling with slow, unstable software releases that exhausted their QA team and eroded trust with the business. They needed a cultural and technical reset — not just better tooling, but better habits.
The Challenge
Release cycles were inefficient and unpredictable, often requiring late-night troubleshooting after deployments went wrong.
QA was stretched thin, the business had lost confidence in the engineering team's ability to ship reliably, and the team had no clear path forward.
The Solution
- GitFlow branching strategy brought structure and consistency to the development process
- Feature flags reduced deployment risk by decoupling releases from code changes
- Increased QA involvement earlier in the cycle caught issues before they reached production
- CI/CD practices formalized the path from commit to deployment
The Results
- Deployment time dropped 83% — from 2+ hours to 20 minutes
- Deployments increased from bi-weekly to 3–4 times per sprint
- After-hours incidents were eliminated entirely
- Three teams have since adopted the same process, with strong praise from the business side
"We now have three of my teams converted to this process and the business can't believe how far we've come in such a short period."
— Senior Director, Digital Engineering
Accelerating the Delivery of a Data Warehouse
An optometry organization needed to unify data from five siloed source systems into a single enterprise data warehouse — built from scratch on a brand-new platform, in 16 weeks.
The Challenge
Four of five source systems were structurally distinct, with no common model between them. One system already conformed to the desired industry-standard design, but the remaining three had to be reverse-engineered and remapped through complex joins and column-level transformations to match it.
The target warehouse didn't yet exist, and the solution had to stay resilient against ongoing schema drift in the source systems — without requiring constant manual rework every time a source changed.
The Solution
- Stood up a Microsoft Fabric warehouse with a config-driven view layer handling schema drift, timezone conversions, PII masking, and soft-delete filtering automatically.
- Built a reusable Type-2 SCD procedure that auto-evolves history tables, generates surrogate keys, and validates completeness on every run.
- Reverse-engineered three source systems into a config-driven Golden Record merging all four sources into a single enterprise version of each dimension.
- Built dimensional key-lookup views for all five fact tables, conforming facts across every source without materialization overhead.
The Results
- Delivered a fully automated, config-driven warehouse architecture in 640 hours — including the learning curve of a new platform
- Conformed 4 structurally distinct sources into a single model across 11 dimensions and 5 fact tables
- Automated daily processing of hundreds of raw tables through two reusable procedures
- Client staff independently onboarded new facts and dimensions post-handover with minimal code changes
"I'd recommend Pinnacle without hesitation to any organization looking for a thoughtful, knowledgeable data architecture partner."
— SVP of IT