Non conformities canevas edition based on Agentic Rag using Dataiku & Svelte.js

This demo is deployed on Github pages

Global Architecture

Global components used :

• Svelte for UI - this repo

  • Non-Confromity Canevas Creation (interactive creation with AI)
  • Contextual information retrieved with RAG (Non-Conformity history, Tech Docs)
  • Deep-Chat for backend-less Chatbot

• Dataiku for AI Agent and dataprep

  • Flask Backend with Agents specialized by task
  • Vector Databases (Chroma DB) for Non Conformity history and Tech docs
  • Tech Doc storage, parsing, chunking and embedding
  • Synthetic data creation for Non-Confomity history (5000 tickets with at least five steps)
  • LLM Mesh for OpenAI embeddings and GPT4o/4o-mini, can be replaced by any-LLM with few clicks

The Agent AI works as follows:

• User provide limited input either on a Non-Conformity task field or directly to the deep-chat input.
• AI Agent looks up to vector databases and provide a json including task fields to update the Non-conformity Edition block (Report or Task 100), reference tech docs and non-conformity tickets (to update left side helpers), and a comment to display within deep-chat
• User can iterate to make the Non-Conformity task fiels more relevant or to translate it to a more comfortable language.

Developing

# install dependencies (legacy option required to use svelte-pdf)
npm install --legacy-peer-deps

# dev mode
npm run dev

Building

To create a production version of your app:

npm run build

You can preview the production build with npm run preview.

Deploy

This repo is using Github Actions to deploy static pages to Github Pages (cf .github/workflows/deploy.yml)