Skip to content

Latest commit

 

History

History
164 lines (154 loc) · 6.26 KB

readme.md

File metadata and controls

164 lines (154 loc) · 6.26 KB
Raw

Work flow

gpt prompt: Please guide me on structuring a secondary backend (Backend 2) that securely handles client requests with JWT tokens and interfaces with an existing LLM API (OpenAI). I want the clients to initiate requests from the front end (thingsboard client) to Backend 2. The client already have a jwt from a backend 1, and will send it to backend 2 to verify validity, and subsequently adapt and forward the main client querie to the LLM API. How should I design this intermediate Backend 2 effectively? Feel free to ask any clarifying questions to help me proceed.

ans: guidances on the concepts and artifacts involved DO: Comprehend ideas Basic

  • Adaptation of user request to conform messages
  • LLM forwarding to API
  • Logging feat for server activities
  • Error handling (invalidations, api errors, etc)
  • Documentation: endpoints, auth, vendors Medium
  • Auth validation for incoming msgs with JWT
  • Security (origin check, token expiration, revocation)
  • Testing: unit, integration, security, api (postman or insomnia) Advanced
  • Scaling: load balancing and cache strategies Further sugestions
  • Rate limiting of usage
  • GraphQL for more felxibility
  • Content security policy
  • Easy to expand endpoints

gpt prompt: Please create the files considering the following

  • An express server to listen from a web client and redirect to OpenAI LLM API
  • Only listen mesages from a client in a DOMAIN_URL env's variable
  • Should validate an incoming string in the msg body, using it as bearer token to ask a BACKEND1 API
  • Server should take a cameraData info in the body and build a prompt for an openai gpt-3.5 turbo model
  • Send a query to the openai gpt-3.5 turbo model using the openai api
  • return the response to the client as chatgptmsg property of the body

IMPORTANT: USE THE BEST PRACTICES TO ORGANICE THE PROJECT. SEPARATE CONCERNS IN A VERY SCALABLE WAY

ans: Guidance on how set up all DO: Comprehend procedure

  • Init project: pnpm init
  • Install dependencies: pnpn install express dotenv
  • Implement custom middelwares
  • Send query to openai
  • Code basic server.js
  • Join functions

gpt prompt: Now please give the vanilla javascript code to build that server. Be organiced and separate concerns in files for further scalability

ans: folder structura and main files DO: build all proposed

  • Folder structure: src(.env + index.js + server.js + /middleware + /routes + /test

FEAT: openai gpt api test

FEAT: Home page serving

  • /public folder with html
  • Indicate express page serving: app.use(express.static(./path/to/public))
  • Learn: all nodejs path are from the root ./

FEAT: Server activities logging

  • Reseacrh logging
  • Install and import morgan
  • Add morgan as express middelware: app.use(morgan('tokenString'))
  • Morgan bases on tokens: app.use(express.static("./src/public"));

EXPERIMENT: Object analyzer function Please help me giving me a JavaScript function to describe a complex Json object in natural language, with enough detail to input the description into an LLM or ChatGPT in order it to later generate a function to traverse the object and apply basic data analytics to that described JSON object.

DUTY: Tight WSL2 memory efficiency on my machine

Set up Digital Ocean sever

  • A cheap server: 1GB, 1core, 25GB

  • as root

    • adduser {username} . Crear usuario
    • usermod -aG sudo {username} . Agregar user al group ENV INSTALLATION

  • curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.7/install.sh | bash . Install nvm (https://github.com/nvm-sh/nvm?tab=readme-ov-file#installing-and-updating)

  • curl -fsSL https://get.pnpm.io/install.sh | sh - . Install pnpm (https://pnpm.io/installation) +

  • nvm install --lts . Install Node LTS

  • pnpm i -g pm2 . Install PM2 RUN PROJECT

    • git clone, pnpm i, node index.js, pm2 logs

  • Some system check

    • top -o %MEM . Sort top by memory usage
    • ps -e . view all active processes
    • ps -e | grep node . check for NodeJS porcesses

ADDED REACT VITE BUILD

  • In other place, loaded created vite to see main files
  • Brougth here the main files an lines
    • vite.config.js in root
    • .jsx files in /src
    • /assets dir in /src
    • .gitignore lines
    • .eslintrc.cjs in root
    • vite.svg file in public (build out)
    • inserted new dependencies in package.json
    • inserted script in package.json
    • run pnpm i
  • Build app with pnpm run build
  • Browser Error: 'Failed to load module script: Expected a JavaScript module script but the server responded with a MIME type of "text/jsx". Strict MIME type checking is enforced for module scripts per HTML spec.'
    • Set JS script path in build's index.html as relative (added '.' before '/')
  • Set the build folder inside inner public
    • Build: { outDir : './path/to/build/dir' } in vite.config.js
  • Error: vite.svg not found
    • Create an dummy/ugly outer public folder in root and added the svg file
    • Seems like docu said: vite consider links as relative, but images as globals
  • Put all this test into a Git Branch

Milestones

  • voice2voice AI assistant beta

    • Capture voice in client
    • Post sound file to backend
    • Forward to whisper and fetch transcript
    • Forward transcript to gpt-turbo and fetch answer
    • Elaborate information corpus
    • Optimize prompting
    • Forward answer to whisper speech
    • Return speech answer to client
    • Play sound in client
    • Show transcript in client

  • Make it pretty

    • Identify TB look and feel ->
    • Interact with gpt to clarify desing procedure
      • Long conversation
      • Main prompt: