Case Study 2 of 7

Healthcare / AI•2025•Live Demo

Vital Healthcare

Clinical Lab Diagnostic Intelligence with Client-Side PII Privacy

My Role

Full-Stack Engineer

Timeline

3 Months

Team

2 Engineers (Aditya: Lead Backend & AI Integration)

Primary Stack

React, Express, Node.js

Visit Live App View Repository

Key Verified Outcomes & Metrics

150+

Lab Tests Parsed

Structured diagnostic indicators mapped

-60%

Report Reading Time

Patients understand lab results 60% faster

100%

Data Masking

Client-side PII redacting before AI transmission

96/100

Accessibility Score

Lighthouse mobile accessibility rating

Project Overview & The Core Problem

Translate complex blood tests and pathology reports into clear visual biomarker trends with client-side privacy guardrails and actionable lifestyle questions for their physician.

My Responsibilities & Contributions

Built the medical report ingestion engine with multi-stage JSON schema extraction using OpenAI GPT-4o.
Developed a HIPAA-conscious client-side sanitization layer that strips Personally Identifiable Information (PII) before LLM analysis.
Created dynamic visual charts for biomarker progression over time (cholesterol, glucose, hemoglobin).
Designed responsive medication reminder workflows with browser notification hooks.

Technical Architecture

System Design & Component Flow

1. Frontend Client

React with responsive Tailwind UI, Recharts for biomarker curves

2. Backend & APIs

Express.js REST API with file upload sanitization and rate limits

3. Database / AI Tooling

MongoDB with encrypted document fields for medical metrics

Data Flow Pipeline

1. Upload

User uploads PDF/photo of lab diagnostic report

2. Sanitize

Client redacts personal identity tokens and phone numbers

3. Extract

AI parses test names, numerical values, and standard reference ranges

4. Visualize

Trends charted against healthy baselines with plain-English takeaways

Engineering Challenges & Implemented Solutions

Building in the real world means encountering unexpected edge cases, latency bottlenecks, and technical trade-offs.

Challenge 01

Unstructured Lab Report Variability

Diagnostic centers output wildly inconsistent PDF formats, table layouts, and unit nomenclatures (e.g., mg/dL vs mmol/L).

Solution Implemented

Strict JSON Schema & Guardrails

Employed OpenAI Structured Outputs with zod schemas strictly defining known reference ranges and fallback warnings.

Challenge 02

Medical Hallucination Prevention

AI summaries must never fabricate diagnoses or prescribe medicines; they must only interpret ranges accurately.

Solution Implemented

Client-Side PII Stripping

Regex and NER filters replace patient names, addresses, and IDs with anonymous tokens prior to external API dispatch.

Technologies Utilized in Vital Healthcare

ReactExpressNode.jsMongoDBTailwind CSSOpenAI

Core Capabilities Shipped

Interactive lab report parser with automated reference range flags (low, normal, high)
Longitudinal health trends visualizing biomarker history across months
Medication schedule tracker with dosage reminders and doctor question prompts
Exportable doctor summary sheet condensing historical tests onto one page

Final Outcome

Demonstrated 60% reduction in time required for non-medical users to understand routine pathology reports with zero safety breaches.