Your Data - Your AI

Materials for the master class "Your Data - Your AI"

Content is grouped by the projects:

• Amazon Killer
• Vibe Analytics
• AI my API

Pre Requisites

• Python 3
• Claude Desktop
• Docker

Amazon Killer

Goal is to create e-shop where you can select and buy products using LLM chat

Step 1 Setting up Database

Checkout the project locally:

git clone https://github.com/bitquery/ydya.git

1. Run MySQL server under docker

 docker run -d \
  --name mysql-amazon \
  -e MYSQL_ROOT_PASSWORD='mysql' \
  -e MYSQL_DATABASE='my_store' \
  -e MYSQL_USER='user' \
  -e MYSQL_PASSWORD='pwd' \
  -p 3306:3306 \
  -v /tmp/mysql:/var/lib/mysql \
  -v /tmp/import:/var/lib/import \
  -v /tmp/mysql-logs:/var/log/mysql \
  --health-cmd='mysqladmin ping -h localhost' \
  --health-interval=10s \
  mysql:8.0-oracle \
  --local-infile=1 \
  --general_log=1 \
  --general_log_file=/var/log/mysql/queries.log \
  --log_output=FILE

Make sure it is started

docker ps
CONTAINER ID   IMAGE              COMMAND                  CREATED         STATUS                   PORTS                               NAMES
a8ffcfa8f2c6   mysql:8.0-oracle   "docker-entrypoint.s…"   3 minutes ago   Up 3 minutes (healthy)   0.0.0.0:3306->3306/tcp, 33060/tcp   mysql-amazon

2. Download and unzip full dataset of products from https://www.kaggle.com/datasets/asaniczka/amazon-products-dataset-2023-1-4m-products

curl -L -o data/amazon-products-dataset-2023-1-4m-products.zip\
  https://www.kaggle.com/api/v1/datasets/download/asaniczka/amazon-products-dataset-2023-1-4m-products

unzip data/amazon-products-dataset-2023-1-4m-products.zip -d /tmp/import

3. Import schema, data into MySQL

go to client of MySQL:

docker exec -it mysql-amazon mysql --user user --local-infile=1 -p    
Enter password: pwd

copy/paste/execute files from data dir:

1. schema.sql
2. import.sql ( can take some time to execute )
3. create_account.sql ( you can replace name and email with your own )

Now you can open file to see queries in other window

tail -f  /tmp/mysql-logs/queries.log

and execute queries in client:

SELECT * FROM my_store.products LIMIT 10;

Step 2 Create code for MCP server

1. Open Claude Desktop, switch to Code tab and connect it to local ydya project

2. Enter the prompt

I have a database created in data/schema.sql for e-commerce. 
Generate MCP server in Python that will allow AI chat to interactwith this database, 
provide recommendations on products and generate orders for them in interactive chat.
Add there methods for full text search of products by words.

3. make sure it created files requirements.txt and server.py

Step 3 Connect LLM to MCP server

1. Open Claude Desktop, go to Settings > Extensions Press button Advanced Settings Press button Install Unpacked Extention

   Select folder of ydya project and install it.

   In case you got an error, check the logs.

   One of the problem can be that it should have a path to correct executable of python.

   To fix it, change "python3" to the correct path of python, for which you already install all required packages

2. Make sure it is started. Extentions now should show ecommerce-mcp extentions with a number of tools

   Press Configure to see the tools.

Step 4 Chat with LLM

1. Go to Claude Desktop in Chat folder and try to buy goods, for example by chat:

I go to Mexico tommorow, fly by plane and want to buy something for this trip, for entertainment and health
I want to buy your list sunscreen cream, sleep mask and cheap Headphones. Propose an order for me
yes, i want to buy these
email is <email>
Send to <address>
yes

2. Ask it to sell the selected goods for you, provide address and make an order.

3. Check MySQL log and order creation

SELECT * FROM my_store.orders;

Step 5 (optional) Debugging tools for MCP

1. Use file mcp.json with MCP inspector

npx @modelcontextprotocol/inspector --config mcp.json

Vibe Analytics

Create application to make analytics of user recommendations using our analytical database on Clickhouse.

Step 1 Create analytical database on Clickhouse

1. Install clickhouse

docker pull clickhouse/clickhouse-server

docker run -d --name clickhouse-server -e CLICKHOUSE_DB=amazon -e CLICKHOUSE_USER=username -e CLICKHOUSE_DEFAULT_ACCESS_MANAGEMENT=1 -e CLICKHOUSE_PASSWORD=password -p 8123:8123 -p 9000:9000 clickhouse/clickhouse-server

2. Import product reviews

docker exec -it clickhouse-server clickhouse-client --user username --password password

copy / paste / execute files

• analytics/schema.sql
• analytics/import.sql

3. Install https://github.com/ClickHouse/mcp-clickhouse

python3 -m pip install mcp-clickhouse

Step 2 Install in Claude Desktop

1. Setup MCP extension in Claude Desktop the same way as we did above, but from the folder analytics.

2. Check that extension is initialized

Step 3 Vibe Analyse

try use propts like:

What analytics can i do with Clickhouse server?
What is the average rating for products by categories?
what is the most rated book ?
which words in the book title causes maximum rating?

AI my API

Step 1 Setup account on Bitquery IDE

1. Go to Bitquery IDE and register
2. Get API key to access

Step 2 Setup account on zuplo service

1. Go to (https://zuplo.com/) and register

Step 3 Create API gateway for GraphQL endpoint from Bitquery

( follow the procedure from the video ...)

Step 4 Connect extension to Claude Desktop as URL and try queries

Step 5 Use inspector to test it

npx @modelcontextprotocol/inspector --server-url 'https://my-api-main-1546619.d2.zuplo.dev/mcp?apiKey=<YOUR KEY>'

Step 6 Connect to the remote MCP server in Claude Desktop using the URL

Do not forget to use full URL with 'apiKey' parameter in query!