Graph AI
Heterogeneous graph neural network for preference-aware entity matching. GNN message-passing over multi-relational interaction graphs with hybrid ranking. Cold-start mitigation across sparse interaction regimes.
Heterogeneous graph neural network for preference-aware entity matching. GNN message-passing learns preference embeddings through neighborhood aggregation. Hybrid ranking combines structural scores with content-based features.
Traditional recommendation relies on aggregate signals. Collaborative filtering suffers from cold-start and data sparsity. The interaction graph contains rich preference signals that flat models ignore.
Heterogeneous graph construction with representation fallback. GNN message-passing learns preference embeddings. Hybrid ranking weighted dynamically based on interaction density. Content-based fallback for cold-start.
Applied AI Engineer
Train and refine PyTorch Geometric models, hybrid ranking, and collaborative filtering integration.
Apply →Backend Engineer
Build graph construction pipeline, Flask APIs, PostgreSQL/Redis serving, and inference integration.
Apply →Data Engineer
Graph construction from interaction logs, feature stores in PostgreSQL, and data quality for GNN inputs.
Apply →AI Researcher
GNN and cold-start experiments, offline evaluation methodology, and hybrid ranking studies.
Apply →DevOps / MLOps
Training and deployment pipelines, Redis caching, monitoring for inference and batch graph builds.
Apply →