This is a submission for the Open Source AI Challenge with pgai and Ollama
What I Built
I created a customer support RAG assistant.
- User request for is submitted (e.g. 'having trouble with XYZ')
- Semantic search is performed to look for similar tickets in DB
- Prompt to LLM includes user query and similar tickets retrieved
- Server returns suggested response to user
Demo
Tools Used
pgvector
Signed up for TimescaleDB service to create a PostgreSQL instance with vector capabilities using pgvector.
Timescale CSV import
Used this dataset from Kaggle with ~600 customer support tickets.
pgai Vectorizer / Open AI API
Created a vectorizer to create vector embeddings of users requests (email body column).
gpt-3.5-turbo
Server calls gpt-3.5-turbo completions API to compose suggested response to user based on their query and similar tickets found.
Source Code
Final Thoughts
Prize categories
Vectorizer Vibeπ
Top comments (0)