Contact person: "WildGuard" Cross-Platform Mobile Application

Phone:Show

Email:Show

Location: Fornebu, Norway

Budget: Recommended by industry experts

Time to start: As soon as possible

Project description:
"Product Specification — “WildGuard” (iOS & Android)

0. We start with iOS app and once working option to develop android. Written partially with a llm so some errors may be present in the requirements.
1. Overview & Goals

WildGuard is a cross-platform mobile application designed to:
• Snap or upload a photo to identify an animal or insect, showing model confidence.
• Indicate if it is dangerous, and in what way (venomous, disease vector, aggressive, invasive, property damage, allergy risk, etc.).
• Verify the sighting’s location using GPS (Mapbox) and EXIF metadata.
• Provide professional, region-specific advice to reduce risk and limit spread, with clear safety and legal disclaimers.
• Contact sponsors (pest control, wildlife officers, municipalities) via call/SMS/email.
• Report rubbish to the responsible authority via prefilled email (with photo, coordinates, timestamp).
• Allow users to create accounts, join leaderboards, and earn recognition for valid reports.
• Allow businesses and farmers to monitor areas of interest (AOI) and receive alerts for potentially harmful sightings.
• Let businesses send “Thanks” messages to reporters.
• Integrate an LLM (e.g., ChatGPT) to:
• Explain dangerous species and safe handling in natural language.
• Answer follow-up questions from users.
• Assist in identifying species when the model confidence is low.
• Ensure all source code is included in the delivery, with a clear contractual clause:
• All intellectual property belongs to the client.
• The code may not be sold, licensed, or shared with any third party without explicit written consent.

Primary Users:
• General public, hikers, gardeners, municipal staff, property managers.

Secondary Users:
• Sponsors, pest-control providers, wildlife authorities, businesses, and farmers.

⸻

2. Platforms & Technology
• Mobile: iOS 15+ (Swift/SwiftUI) and Android 9+ (Kotlin/Jetpack Compose)
OR cross-platform (Flutter or React Native).
• Maps & Geocoding: Mapbox Maps SDK + Mapbox Geocoding (forward/reverse), optional geofencing.
• ML Inference:
• On-device: Core ML (iOS), TensorFlow Lite/NNAPI (Android).
• Cloud fallback: REST API with GPU/CPU autoscaling.
• LLM Integration: ChatGPT API (or equivalent) with:
• Local caching of Q&A to reduce calls.
• Guardrails to prevent unsafe or illegal advice.
• Storage: Cloud object storage for images (signed URLs), Postgres DB (sightings, taxonomy, users, AOIs), Redis (cache).
• Notifications: FCM (Android), APNs (iOS).
• Analytics/Logs: Privacy-compliant telemetry; crash reporting.

⸻

3. Core Features

3.1 Capture & Upload
• Capture via camera or select from gallery.
• Parse EXIF for GPS, timestamp, and camera data.
• If EXIF GPS exists and user consents → prefill location; else use device GPS; else manual pin via Mapbox.
• Clear permission prompts for camera, photos, and location.

3.2 Identification
• On-device model runs first; if confidence < threshold or species is out-of-distribution (OOD) → cloud model.
• Display:
• Common name + scientific name.
• Confidence percentage + top 3 predictions.
• Short explanation (from model and/or LLM) for why it matches.
• Danger badge: Safe / Caution / Dangerous / Invasive.
• Danger details: mechanism, severity, first aid, seasonality.
• Local status: native, protected, invasive.
• Prevention & control advice.

3.3 Location Handling
• Auto GPS, manual pin, and address search.
• Show accuracy, timestamp, and location source.
• Map overlays for:
• Sightings heatmap.
• Rubbish reports.
• Risk zones.
• Optional offline tile packs.

3.4 Verification via EXIF
• Compare EXIF and device GPS.
• Flag mismatches and allow user to choose.

3.5 Professional Advice
• Region-specific guidance (via database + LLM refinement).
• Containment, legal notes, IPM steps for insects.
• Direct “Contact a professional” link.

3.6 Contact Sponsors / Authorities
• Call, SMS, or email nearest sponsor or municipal contact.
• Prefilled email includes:
• Species name, risk, map link, coordinates, timestamp, photo link, user notes.
• One-tap emergency call.

3.7 Report Rubbish
• Quick-access report flow.
• Attach photo, location, category, size, notes.
• Send to authority via email.

3.8 Confidence & Explainability
• Confidence meter with color thresholds.
• LLM explanations for low-confidence identifications.
• Suggest actions for improved identification.

⸻

4. Accounts, Leaderboards & Business Features

4.1 Accounts & Identity
• Auth via email/password or Apple/Google sign-in.
• Profiles with handle, avatar, badges, privacy settings.
• Roles: User, Business, Farmer, Sponsor, Admin.

4.2 Leaderboards & Reputation
• Points for verified reports:
• Animal/pest: 10 pts (+5 for dangerous/invasive).
• Rubbish: 5 pts (+3 if marked resolved).
• Quality multipliers for verified GPS, high confidence, or validations.
• Anti-cheating measures (pHash, clustering, caps).

4.3 Business & Farmer Accounts
• AOI creation (up to N per plan) with geometry drawing or GeoJSON import.
• Heatmaps and alerts for dangerous species in AOI.
• Data export and daily/weekly digest.
• Send “Thanks” messages to reporters.

4.4 Kudos System
• Verified orgs can send short text “Thanks” messages.
• Reporter notified in-app and via email.
• Points awarded for received thanks.

⸻

5. LLM Integration Details
• LLM provides:
• Plain-language summaries of identification results.
• Clarifications on danger and safe handling.
• Suggestions for better image capture for re-identification.
• Safety:
• Pre-filter all LLM outputs through a safety layer.
• Block advice that contradicts legal or medical guidelines.
• Offline fallback for basic predefined responses.

⸻

6. Security, Compliance & Licensing
• All data encrypted in transit (TLS) and at rest (AES-256).
• GDPR compliance: consent management, opt-out, right-to-be-forgotten.
• Role-based data access; location blurring for privacy.
• Licensing clause:
• All source code, models, training scripts, and assets are delivered to the client.
• Ownership remains solely with the client.
• Vendor/developers may not sell, license, or share the code or derivatives with any third party without explicit written consent.

⸻

7. Non-Functional Requirements
• Performance: On-device inference p95 < 800ms; cloud inference < 3s.
• Reliability: Offline capture + queue for later sync.
• Accessibility: WCAG 2.2 AA compliance.
• Localization: Multi-language UI & species data.


⸻" (client-provided description)