Multimodal AI & Structured Systems
PetAI
Turn pet images, symptoms and documents into structured guidance.
PetAI helps pet owners evaluate food safety, organise veterinary records and understand potential concerns through multimodal AI. It provides informational guidance and follow-up support, not veterinary diagnosis.
- Role
- Solo Product Engineer
- Platform
- iOS
- Scope
- Flutter development, TypeScript backend services, multimodal AI, structured data extraction, validation and asynchronous workflows
Pet information is often fragmented across photos, food labels, symptom descriptions and veterinary documents.
The challenge was to turn these mixed inputs into consistent and inspectable product data without treating model-generated responses as inherently reliable.
Input
Add an image, symptom description, food item or veterinary document.
Analysis pipeline
- Input validation and sanitisation
- Multimodal analysis
- Structured JSON generation
- Schema validation
- Firestore persistence
- Contextual follow-up
Experience
Previous pet records and completed analyses can be used as context for later conversations instead of treating each request as an isolated prompt.
Three implementation choices that made the system dependable.
Validated structured output
AI responses are transformed into predefined JSON structures and checked against schemas before being stored or presented.
This prevents unvalidated free-form model output from becoming trusted application data.
Modular TypeScript services
Health analysis, nutrition checks, document extraction, contextual chat and media generation were separated into dedicated backend services.
The modular structure kept unrelated AI workflows from accumulating inside a single Cloud Function.
Asynchronous media jobs
Long-running image-to-video requests use job IDs, status polling, retries, timeouts and Firebase Storage persistence.
This allows the mobile client to track work that cannot be completed inside a single synchronous request.
Released on the App Store and validated through initial purchases and recurring subscriptions.
- 3K+
- App Store impressions
- 23
- In-app purchase events
- 2
- Active paid plans
- First
- Recurring revenue
The technical system supported several strong use cases, but the product scope became too broad.
Health guidance, document organisation and nutrition checks belonged to a trust-focused product, while generative entertainment features created a separate identity.
A stronger next version would narrow the product around pet records, food-safety guidance and contextual follow-up.