law cn ai

AI Legal Assistant

Legal document source: https://github.com/LawRefBook/Laws

Project template: https://github.com/supabase-community/nextjs-openai-doc-search

This project takes all .mdx files from the pages directory and processes them into a custom context for use in OpenAI text autocomplete prompts.

More fun

Smart Writing Assistant	AI Encyclopedia	Prompt generator
AI Translation Expert	❤️ Reward and sponsorship ❤️

deploy

Deploy this starter to Vercel. Supabase integration automatically sets the required environment variables and configures your database profile. You just need to set OPENAI_KEY and you're ready to go!

The poster is too lazy, thanks to GoJun for helping to write the tutorial: https://eibot3u32o.feishu.cn/docx/L46Pdp3fjouPUvxaNzPckKctno3

Technical details

Building your own custom ChatGPT involves four steps:

1. [? Build Time] Preprocessing knowledge base (the .mdx file in your pages folder).
2. [? Build Time] Use pgvector to store embedded vectors in PostgreSQL.
3. [? Runtime] Perform a vector similarity search to find content related to the problem.
4. [? Runtime] Inject content into OpenAI GPT-3 text autocomplete and stream responses to the client.

? Build time

Steps 1 and 2 occur at build time, such as when Vercel builds your Next.js application. At this time, the generate-embeddings script is executed, which performs the following tasks:

 sequenceDiagram
    participant Vercel
    participant DB (pgvector)
    participant OpenAI (API)
    loop 1. Preprocessing knowledge base Vercel->>Vercel: Divide the .mdx page into parts loop 2. Create and store embed Vercel->>OpenAI (API): Create embed OpenAI (API) for the page part->>Vercel: embed vector (1536)
            Vercel->>DB (pgvector): Embed end of the page part
    end

In addition to storing the embed vectors, this script generates a checksum for each .mdx file and stores it in another database table to ensure that the embed vectors are regenerated only if the file changes.

? Runtime

Steps 3 and 4 occur at runtime, i.e. when the user submits the problem. When this happens, perform the following series of tasks:

 sequenceDiagram
    participant Client
    participant Edge Function
    participant DB (pgvector)
    participant OpenAI (API)
    Client->>Edge Function: { query: lorem ispum }
    critical 3. Perform vector similarity search Edge Function->>OpenAI (API): Create an embedded OpenAI (API) for a query ->>Edge Function: Embed vector (1536)
        Edge Function->>DB (pgvector): Vector Similarity Search DB (pgvector)->>Edge Function: Related Document Content End
    critical 4. Inject content into the prompt Edge Function->>OpenAI (API): Complete request prompt: Query + related document content OpenAI (API)-->>Client: text/event-stream: Automatic completion of response end

This is the relevant file responsible for SearchDialog（客户端） component and vector-search（边缘函数） .

The initialization of the database, including the settings for pgvector extension, is stored in supabase/migrations folder and is automatically applied to the local PostgreSQL instance when supabase start is run.

Local development

Configuration

• cp .env.example .env
• Set OPENAI_KEY in the newly created .env file.

Start Supabase

Make sure Docker is installed and running locally. Then run

npx supabase start

Start the Next.js application

Run in a new terminal window

pnpm dev

deploy

Just deploy this starter to Vercel. Supabase integration automatically sets the required environment variables and configures your database Schema. You just need to set OPENAI_KEY and get started!

learn more

• Read our blog post on How We Build Supabase Documentation ChatGPT.
• [Docs] pgvector: Embed and vector similarity functions.
• Watch Greg's "How I built this" video about Rabbit Hole Syndrome YouTube Channel:

此文件由 ChatGPT 提供翻译

Star History