Workshop: Vector Search with LangChain and MariaDB

Learn how to use MariaDB as a vector store (vector database) for embeddings using Gemini, Python, and LangChain.

---

➡️ Step 1 - Download Docker version 28 or later and run the following in a terminal window:

docker --version

---

➡️ Step 2 - Download (click on Code -> Download ZIP) and extract the project at:

https://github.com/mariadb-developers/vector-search-workshop

Note: Alternatively, you can clone the project using Git.

---

➡️ Step 3 - In a terminal window, move to the project directory, and start the services with:

docker compose up -d

---

➡️ Step 4 - Watch the logs with:

docker logs code-server -f

---

➡️ Step 5 - Go to http://localhost:4444/ and create a MariaDB connection (click on Database icon -> Create Connection -> MariaDB):

• Host: mariadb-server
• Port: 3306
• User: root
• Password: RootPassword123!

---

➡️ Step 6 - Click on the Open Query button next to demo and run this SQL query (replace the search query first!):

SELECT CONCAT(ROW_NUMBER() OVER(), ". ", name, " - ", description)
FROM products
WHERE MATCH(description) AGAINST ("YOUR_QUERY_HERE")
LIMIT 10;

---

➡️ Step 7 - Click on Toggle Panel -> TERMINAL and calculate the vectors by running the following commands (replace the API key first!):

export GEMINI_API_KEY=YOUR_GEMINI_API_KEY_HERE
python ingest_embeddings.py

---

➡️ Step 8 - Find products using vector search by introducing your own search query in the semantic_search.py program (use the same search query as before!):

results = vectorStore.similarity_search("YOUR_QUERY_HERE", k=10)

... and running:

python semantic_search.py

---

➡️ Step 9 - Compare results using an LLM (replace the search query and result sets first!):

Which result set is better for the search YOUR_QUERY_HERE" and why? (answer in a very short paragraph)

Result set 1:
INSERT_RESULT_SET_1

Result set 2:
INSERT_RESULT_SET_2

---

➡️ (Optional) Step 10 - Stop and delete services and volumes (this deletes all the data!):

docker compose down -v