The latest articles on DEV Community by kobtze (@kobtze). https://dev.to/kobtze https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F832275%2Fee15416e-9607-488d-bb0a-076e56abfbe5.png https://dev.to/kobtze en kobtze Mon, 11 Nov 2024 13:09:00 +0000 https://dev.to/kobtze/customer-support-rag-assistant-1kka https://dev.to/kobtze/customer-support-rag-assistant-1kka <p><em>This is a submission for the <a href="https://dev.to/challenges/pgai">Open Source AI Challenge with pgai and Ollama </a></em></p> <h2> What I Built </h2> <p>I created a customer support RAG assistant.</p> <ol> <li>User request for is submitted (e.g. 'having trouble with XYZ')</li> <li>Semantic search is performed to look for similar tickets in DB</li> <li>Prompt to LLM includes user query and similar tickets retrieved</li> <li>Server returns suggested response to user</li> </ol> <h2> Demo </h2> <p><a href="https://bit.ly/4flG7xI" rel="noopener noreferrer">https://bit.ly/4flG7xI</a><br> <a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fp0lcqb2aipqyybuasn9x.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fp0lcqb2aipqyybuasn9x.png" alt="MS Office issue" width="574" height="386"></a></p> <h2> Tools Used </h2> <h3> pgvector </h3> <p>Signed up for TimescaleDB service to create a PostgreSQL instance with vector capabilities using pgvector.</p> <h3> Timescale CSV import </h3> <p>Used <a href="https://www.kaggle.com/datasets/tobiasbueck/multilingual-customer-support-tickets" rel="noopener noreferrer">this</a> dataset from Kaggle with ~600 customer support tickets.</p> <h3> pgai Vectorizer / Open AI API </h3> <p>Created a vectorizer to create vector embeddings of users requests (email body column).</p> <h3> gpt-3.5-turbo </h3> <p>Server calls gpt-3.5-turbo completions API to compose suggested response to user based on their query and similar tickets found.</p> <h2> Source Code </h2> <p><a href="https://github.com/kobtze/cs-rag-server-web-app" rel="noopener noreferrer">Server App</a><br> <a href="https://github.com/kobtze/cs-rag-client-web-app" rel="noopener noreferrer">Client Web App</a></p> <h2> Final Thoughts </h2> <h3> Prize categories </h3> <p>Vectorizer Vibe😃</p> devchallenge pgaichallenge database ai