Full-Stack Platform
Gaudia
Queue Management System
TypeScriptReactNode.jsGraphQL+3
View details Hey, I'm Pietro Maltese.Friends Everyone calls me Piero.I'm a Software Engineer.
Selected Work
A collection of projects I've built.
AI / SaaS
Ocarina
AI Document Processing
HonoReactPostgreSQLOpenAI+3
View details PWA / Enterprise
Cadidro
Field Management System
ReactHonotRPCPostgreSQL+3
View details About
London-based full-stack engineer who loves shipping products end-to-end.
Along the way, I've been lucky to work with excellent professionals who fueled the passion I've always had for computing and programming.
I've hopped between backend and frontend, built mobile experiences, explored UI/UX and embedded systems, and spent time with realtime systems, SQL/NoSQL data stores, microservices, event-driven architectures, serverless, and observability.
Get in touch
AI / SaaS
Ocarina
AI Document Processing
Ocarina automates the extraction of structured data from unstructured documents like contracts, invoices, and forms. The platform uses a multi-stage pipeline combining OCR, AI analysis, and custom attribute extraction to transform documents into actionable data.
HonoReactPostgreSQLOpenAIAnthropicAWS TextractPrisma
Screenshots
Key Features
- Multi-provider AI processing (OpenAI, Anthropic Claude, Google Gemini)
- AWS Textract integration for OCR and text extraction
- Custom attribute groups with JSON Schema definitions
- Document management with status tracking
- Background job processing with Inngest
- Batch processing for bulk document uploads
- RESTful API for third-party integrations
- Multi-user support with audit logging
Technical Implementation
Backend
Hono framework with type-safe routing on Node.js
Database
PostgreSQL with Prisma ORM
AI Providers
Supports latest models from OpenAI, Anthropic, and Google
Background Jobs
Inngest for reliable job processing and retries
Frontend
React 19 + Vite with TanStack Router and Query
Deployment
Serverless on Vercel with edge functions
Challenges
Handling varying document quality and formats required building a flexible processing pipeline that could gracefully handle OCR failures and route documents to the most appropriate AI model based on complexity.
Outcome
The platform handles the full extraction lifecycle, from raw scan to structured output, with confidence scoring that flags uncertain fields for human review, removing the need for manual data entry on well-formatted documents.