Hierarchical Classifier

Demo

Overview

During my AI research internship at Utah Tech University, I implemented hierarchical classification using vector embeddings and similarity search. The project combines a taxonomy import pipeline, a PostgreSQL database, and a FastAPI service for classifying inputs within a selected hierarchy.

Role: AI Research Intern

Technical Stack

• •Python
• •FastAPI
• •PostgreSQL / pgvector
• •OpenAI embeddings
• •AWS Elastic Beanstalk and RDS

Notable Features & Challenges

• 1Import transformers for hierarchical data represented by nodes, parents, and descriptions
• 2Semantic embeddings and similarity search for taxonomy classification
• 3An API for selecting the classification hierarchy
• 4Performance profiling and analysis of classification results

Outcome & Status

The project README documents a historical peak of 49.5 requests per second using seven t3.micro application instances and text-embedding-3-large. The repository includes profiling results and classification analysis alongside the implementation.