NovaGuard

Inspiration

Healthcare providers, financial institutions, and legal firms face a critical dilemma: they need AI to process sensitive documents, but can't risk sending patient records, financial data, or confidential contracts to third-party AI services that retain data for training.

I built NovaGuard to solve this: a privacy-first multimodal AI that analyzes medical prescriptions, financial documents, and sensitive images without ever storing them - using Amazon Nova's powerful multimodal capabilities via AWS Bedrock.

What it does

NovaGuard allows users to:

Upload sensitive images (prescriptions, invoices, contracts, medical scans)
Get AI-powered analysis using Amazon Nova Pro's vision + reasoning
Receive actionable insights in plain language
Guaranteed zero data retention - images processed in-memory, never saved to disk

Key Workflows:

Medical Use Case: Upload prescription image -> Nova extracts medications, dosages, instructions -> Summarizes in patient-friendly language -> Image immediately discarded

Financial Use Case: Upload invoice/receipt -> Nova extracts line items, totals, vendor info -> Categorizes expenses -> Data never persisted

Legal Use Case: Upload contract page -> Nova identifies key terms, obligations, dates -> Summarizes risks -> No cloud storage

How we built it

Tech Stack: Node.js + Express, Amazon Nova Pro via AWS Bedrock, in-memory processing

Amazon Nova Integration:

Amazon Nova Pro (amazon.nova-pro-v1:0) - Unified multimodal vision + reasoning in a single API call
AWS Bedrock Converse API - Enterprise-grade ephemeral processing with built-in HIPAA/SOC 2 compliance
In-Memory Architecture - Multer memory storage, zero disk writes, garbage collection clears buffers

Why Amazon Nova?

Unified multimodal model - vision + reasoning in ONE API call (vs separate models)
AWS Bedrock compliance - built-in HIPAA, SOC 2, ISO certifications
Ephemeral processing - no data retention by design
Enterprise security - IAM, VPC, regional deployment

Challenges we ran into

Handling diverse image formats (JPEG/PNG/GIF/WebP) with proper MIME type detection for Nova's Converse API
Balancing privacy with functionality - deliberately removed user accounts, history, saved results
Making privacy tangible in a demo without compromising the zero-retention architecture

Accomplishments that we're proud of

Built something that solves a real enterprise compliance problem
Amazon Nova Pro handles vision + reasoning in a single call - incredibly efficient
Zero technical debt for privacy - no database, no disk writes, no logs
Clean architecture that could be deployed in regulated environments today

What we learned

Amazon Nova via AWS Bedrock is a game-changer for privacy-critical AI applications. The Converse API's unified multimodal approach eliminates the need for complex model orchestration, and Bedrock's compliance certifications make it production-ready for regulated industries.