ML.NET Similarity Builder

This tool provides a structured approach to building ML.NET similarity models for SQL Server data. It targets data engineers, data scientists, and analysts who need scalable, auditable scoring integrated into SQL workflows. It enables ranking, clustering, and record linkage by mapping heterogeneous features into a common vector space. The system emphasizes reliability, reproducibility, and compatibility with existing SQL Server processes.
Conceptually, it ingests a user-defined schema, applies feature engineering (normalization, embeddings, numeric bucketing), and trains a similarity model using ML.NET. Users can select metrics such as cosine, euclidean, or Manhattan distances, adjust embedding dimensions, and evaluate with held-out data. The resulting scoring pipeline can be deployed to SQL Server, allowing in-database similarity lookups on large datasets.
Benefits include improved search, deduplication, and content-based grouping, with tight integration into SQL workflows. It provides versioned model artifacts and a clear export path for production scoring. Typical use cases include document similarity, product matching, and entity resolution in enterprise stores.
Core features cover input schemas, metric selection, vector feature creation, model export, and batch scoring. Advanced users can enable cross-validation, incremental updates, and hyperparameter tuning. The tool favors transparency by exposing feature processing steps and maintains logs for traceability.