I built a system to analyze and rank unstructured text using semantic similarity and structured scoring logic. It embeds documents, compares them against both a curated set of positive example documents and keyword-based embeddings, and evaluates relevance using multiple matching strategies with configurable weights and thresholds. It also has an exclusion step where it compares documents to a set of negative example documents. The system is modular, configurable, and runs efficiently in batch.
Built with
- Machine Learning
- Vector Database
- Python