GloxPaw AI

Main banner

Inspiration

We were sitting in a Discord call watching friends post photos of their pets asking "is my dog okay?" or "why is my cat acting weird?" and realized there's a massive gap between casual pet owners and professional veterinary care. People often don't know when to be concerned or what to actually look for.

We thought: What if there was an instant second opinion tool powered by AI that could teach people how to observe their animals better? Not replacing vets—helping people understand what they're actually seeing.

One of us had a rescue cat with behavioral issues and wished they had caught the stress signals earlier. That's when GloxPaw was born.

What it does

GloxPaw AI analyzes any animal photo and generates a comprehensive welfare report that includes:

Species & breed identification with confidence levels
Behavioral analysis - posture, ear position, eye contact, body tension
Physical observations - coat condition, body composition, visible characteristics
Welfare assessment - positive indicators, concerns to watch, what can't be assessed
Environmental context - setting, hazards, enrichment
Actionable recommendations - immediate actions, ongoing observation, when to see a professional

The key: It's not pretending to diagnose. Every observation includes confidence levels and clear disclaimers about what a photo can't tell you.

How we built it

Tech stack:

Python 3.8 - Clean, readable backend
OpenRouter API - Flexible multi-model vision AI integration
Pillow - Image processing and optimization
python-dotenv - Secure configuration

Architecture:

User uploads animal photo
Image validation & Base64 encoding
Send to OpenRouter with comprehensive analysis prompt
Parse structured JSON response from AI
Format into beautiful, welfare-focused report
Save to outputs/

Key decisions:

Used OpenRouter instead of single API provider for flexibility
Structured prompts to ensure consistent JSON responses
Confidence-based reporting (not definitive claims)
Extensive error handling and user validation

Development process:

Started with basic breed classifier
Evolved into welfare-focused assessment tool
Added behavioral + environmental analysis
Built comprehensive documentation & setup verification

Challenges we ran into

1. Image Encoding & API Integration

Struggled getting OpenRouter to properly receive base64-encoded images
Had to debug the exact image_url format OpenRouter expects
Solution: Switched from custom image format to standard data:image/jpeg;base64 with proper message structure

2. AI Response Inconsistency

Models weren't always returning JSON—sometimes text responses
Built fallback parsing for non-JSON responses
Added type checking throughout report generation

3. Balancing Accuracy vs. Responsibility

Tempting to make predictions sound certain, but that's dangerous with animal welfare
Had to find language that's confident but honest about limitations
Created clear disclaimer sections that don't feel like legal fine print

4. Model Selection

Vision capability varies wildly across models
Text-only models can't see images at all
Had to test multiple models to find affordable options that actually work
Cost optimization was crucial for user accessibility

5. Error Messages

Users were getting cryptic API errors
Added debugging output and verification script
test_setup.py now catches 90% of config issues before runtime

Accomplishments that we're proud of

✅ Built a working product in 48 hours - Full stack from concept to deployment-ready code

✅ Welfare-focused approach - Designed the tool to prioritize animal wellbeing over flashy AI claims. Every feature has proper disclaimers.

✅ Clean, production-ready code - Not hacky. Well-structured, documented, error-handled, extensible.

✅ Comprehensive documentation - 30KB+ of guides, troubleshooting, model selection. QUICKSTART.md gets you running in 5 minutes.

✅ Multi-model flexibility - Works with GPT-4, Claude, Mistral, etc. Users can swap models in one line of config.

✅ Setup verification - test_setup.py validates environment, dependencies, API connection before you waste time debugging.

✅ Beautiful reports - Not raw JSON. Formatted, emoji-enhanced, actionable welfare reports that feel professional.

✅ Affordable - Works with models as cheap as $0.001/analysis. Anyone can use it.

✅ Open source ready - Full GitHub package with MIT license, proper .gitignore, professional README, badge integration.

What we learned

Technical:

Vision AI capabilities vary wildly—not all models can handle images
Base64 image encoding requires exact API format compliance
Error handling is 50% of user experience for APIs
Structured prompts (JSON requests) are way more reliable than text

Product:

People want this. Animal welfare is deeply personal and underserved by tech
Confidence levels matter more than perfect accuracy
Disclaimers can be user-friendly if done right
Documentation quality = adoption rate

Design:

Welfare-focused messaging resonates way better than "cool AI" hype
Clear limitations build trust (not the opposite)
Users value honest "can't assess" answers over guesses
Beautiful formatting makes technical output feel accessible

Process:

Build verification tools early (saved us debugging headaches)
Test with real users ASAP (even just Discord friends)
Write docs as you build (not at the end)
Open source mindset from day 1 makes everything cleaner

What's next for GloxPaw AI

Immediate (v1.1):

[ ] Batch image processing - analyze 50 photos at once
[ ] Multiple animal comparison - track same animal over time
[ ] Custom prompt templates - users can define their own analysis focus
[ ] Export to PDF/CSV - integrate reports into care logs

Short-term (v2.0):

[ ] Web interface - No command line needed, just upload photo
[ ] Mobile app - iOS/Android for field use by rescues and shelters
[ ] Shelter integration - Bulk processing for rescue operations
[ ] Real-time monitoring - Video analysis for behavioral tracking

Medium-term:

[ ] Multilingual support - Reach global animal welfare community
[ ] Community database - Anonymized breed/behavior patterns
[ ] Professional tools - API for veterinarians and behaviorists
[ ] AI fine-tuning - Custom models trained on animal data

The dream: We want GloxPaw to be the standard tool for smart observation—helping millions of people understand their animals better. From casual pet owners to professional rescues, everyone wins when we catch problems early.

Long-term: Partner with animal organizations, integrate into veterinary workflows, make animal welfare assessment accessible to everyone.

We're open to: