In this workshop in the Drupal Mountain Camp 2025 (https://drupalmountaincamp.ch/schedule-2025), you will learn how to build the most basic AI supported search / chatbot. We will crawl a bunch of pages and then build a small chatbot against its content.
You will learn the most important technical terms like Retrieval-Augmented Generation (RAG), vectorization, chunking and learn more about using contexts and writing meaningful prompts.
I will give a short introduction and explain the basics about how AI search / chatbots work. Then we will build a small chatbot during the talk using TypeScript and a bunch of libraries. There will be a public repo and all participants will be able to follow and try it out. To sum up, I will talk about the challenges and pitfalls we faced in the last two year when building large AI based chatbots and searches like Alva, the AI search for Kanton Basel Stadt: https://www.bs.ch/alva
Open the URL in the browser and start coding:
https://gitpod.io/new/#https://github.com/ayalon/mini-chatbot
git clone git@github.com:ayalon/mini-chatbot.git
cd mini-chatbotnpm installdocker compose up -dThe database will be available at localhost:5432 with the following
credentials: postgres/postgres
cp .env.example .env
npx drizzle-kit migrateHead over to the src folder and start implementing the 5 steps:
- Crawl the pages:
npx tsx src/01_crawl.ts - Fetch the HTML content:
npx tsx src/02_fetch_html.ts - Chunk the pages:
npx tsx src/03_chunking.ts - Create the embeddings:
npx tsx src/04_create_embeddings.ts - Build the chatbot:
npm run chat
Create DB Dump
docker exec mini-chatbot-db-1 pg_dump -h localhost -p 5432 -d mini_chatbot -U postgres -s -F p -E UTF-8 > dump.sql