NGO: Python Backend Developer for Machine Learning API

Description: We are a registered non-profit organization (NGO) operating Mascotafind (www.mascotafind.es), the largest free platform in Spain dedicated to registering, locating, and saving lost and found pets. Our mission is entirely humanitarian and focused on animal welfare. To scale our impact and reunite families with their pets faster, we want to upgrade our platform with AI. We are looking for a socially-minded, experienced Computer Vision / Machine Learning Engineer to help us build a custom backend pipeline for automated pet identification based on photos. As a non-profit, we have a limited budget, but we offer a highly meaningful project, a great addition to your portfolio, and deep gratitude from the animal welfare community. Architecture Context & Cost Constraints: • Our platform is built on TYPO3 CMS. The AI system must run as an independent, external backend service and communicate with our TYPO3 system via a lightweight REST API. • Because we are an NGO funded by donations, minimizing ongoing server/cloud costs is critical. The architecture must utilize open-source models and cost-effective vector database tiers (ideally free tiers or very low-cost options like Qdrant/Pinecone serverless) so that running costs remain close to zero for our current scale. • Future Expansion: In the next phase, this system should be scalable enough to interconnect with external databases from various animal shelters (refugios) across Spain, allowing cross-platform image matching. What we do NOT want: • No breed classification (we don't need to know the breed). • No muzzle/nose-print scanning (we already evaluated this). • No frontend registry development (our TYPO3 database and website are already fully operational). What we NEED: An automated image-to-image similarity search engine (1:N Matching). When a user uploads a photo of a found pet on our TYPO3 site, the system must analyze the visual characteristics (fur patterns, facial landmarks, colors) and instantly query our database of lost pets to return the top 5 most visually similar matches. Scope of Work / Key Technical Requirements: 1 Object Detection & Cropping: Automatically detect the animal (dog/cat) in the uploaded image and crop the background using open-source models like MegaDetector or YOLOv8/v10. 2 Feature Extraction (Embeddings): Pass the cropped image through a pre-trained Vision Transformer (ViT) or a Deep Metric Learning model (e.g., ArcFace/CosFace adapted for animals) to generate a unique mathematical vector (embedding) for the pet's appearance. 3 Vector Database Integration: Store and query these embeddings using an efficient vector database (e.g., Pinecone, Qdrant, or Milvus) to enable real-time similarity matching. 4 Metadata Filtering: The vector search must support pre-filtering based on simple metadata (e.g., matching only within specific Spanish regions/provinces passed from TYPO3). 5 API Backend: Deliver this pipeline as a lightweight, secure, and well-documented REST API (preferably Python/FastAPI) packaged in a Docker container for easy deployment. Your Profile: • Proven track record in Computer Vision, specifically in Content-Based Image Retrieval (CBIR) or Re-Identification (Re-ID) systems. • Strong experience with Python, PyTorch/TensorFlow, and OpenCV. • Experience with Vector Databases and deploying ML models to production environments (AWS, GCP, or similar). Please provide examples of similar image matching or object re-identification projects you have successfully built in the past. To ensure you read the full description, please start your proposal with the word "Mascota".