BYOD: Vector Databases for AI-driven Web Apps!

At the end of this workshop, you too will have an idea of how to go about building AI-enabled experiences over YOUR data with the latest and greatest from Google models.

Given at DevFest Stockholm

Disclaimer: You will have to setup a billing account to use Google Models. Worry not if you don't have that, you can use something else.

Hello Sweden!

• Repo
• Slides

What are we building today

An AI-powered personal assistant powered by our journal entries.

We'll use...

• Weaviate
• Next.js
• Google (maybe)

Test Setup

Please have the following installed:

• Node.js (v18+): node -v
• VSCode (optional): code .
• Google Cloud CLI (optional): gcloud auth print-access-token - expires in an hour.

Google specifics

1. Setup a project and get our API keys
2. Add Billing to your account
3. Install Google Cloud CLI

Project ID can be found here.

Now let's chat with Google

Let's talk with Google! Go here.

What data are we working with?

Some journal data! Month entries to give as context to our LLM

• Link to data: Journal entries

Test Weaviate connection

1. Make a new folder and run npm init in it.
2. Install weaviate-client, typescript and dotenv.
3. Get credentials on Weaviate Cloud
4. Setup TS config
5. Create .env file.
6. Make a new script file and run it, but before!

Paste this!

  const weaviateURL = process.env.WEAVIATE_URL as string
  const weaviateKey = process.env.WEAVIATE_ADMIN_KEY as string
  const googleKey = process.env.GOOGLE_API_KEY as string

  const client: WeaviateClient = await weaviate.connectToWeaviateCloud(weaviateURL, {
    authCredentials: new weaviate.ApiKey(weaviateKey),
    headers: {
      'X-Goog-Vertex-Api-Key': googleKey,  
    }
  })

  console.log("Client details", (await client.getMeta()).version)

Importing data

1. Create a file called load.ts
2. Let's import data with insertMany()!
   • Data link
3. Run code!

Searching for Data

1. Create a file called query.ts
2. Play with Search using nearText()
3. Run code!
4. What do our vectors look like?

RAG! Retrieval Augmented Generation

1. Let's play with generate.nearText()
2. Prompting 🪄
3. Different types of context! Grouped or Single

Better Search

How can we improve retrieval?

Hybrid Search

The best of both worlds!

1. Alpha and Fusion algorithms

Filters and Aggregates

Can we drill down into our data?

1. Filtering on properties
2. Aggregate queries

Building our App

1. Clone - https://github.com/weaviate-tutorials/nextjs-weaviate-template
2. Run yarn install and yarn dev
3. What are we looking at?
4. Let's make it our own!

Resources

• Weaviate Academy
• Weaviate x JavaScript

Definitions

Refer to slides! Weaviate Knowledge cards help a lot too!