DocChat (formally ChatApp)

About

This repo contains a React front-end application, and a Node back-end application.

A user can upload a PDF file (in English) and then ask a questions about the document in any language.

• The file is stored (and retrieved) in Azure AI Search
  • Embeddings are additionally stored and form part of the hybrid search for RAG content retrievel
• Azure OpenAI (gpt-35-turbo-16k and text-embedding-ada-002) is used
  • To translate any question into English for search
  • To generate embeddings
  • To deduce answers based on search results

AI Search index

An Azure AI Index is required. Free tier is fine for testing.

An AI Search index should be set up with the following schema:-

{
    "name" : "chat-comp-index",
    "fields": [
        { "name": "Id", "type": "Edm.String", "key": true, "searchable": true },
        { "name": "SessionId", "type": "Edm.String", "key": false, "searchable": true, "filterable": true },
        { "name": "ChunkId", "type": "Edm.String", "key": false, "searchable": true, "filterable": true },
        { "name": "FileName", "type": "Edm.String", "key": false, "searchable": true, "filterable": false, "sortable": false, "facetable": false },
         { "name": "FileSize", "type": "Edm.Int32", "key": false, "searchable": false, "filterable": false, "sortable": false, "facetable": false },
        { "name": "Text", "type": "Edm.String", "key": false, "searchable": true, "filterable": false },
        {
            "name": "Embedding",
            "type": "Collection(Edm.Single)",
            "searchable": true,
            "retrievable": true,
            "dimensions": 1536,
            "vectorSearchConfiguration": "vectorConfig"
        }
    ],
        "vectorSearch": {
        "algorithmConfigurations": [
            {
                "name": "vectorConfig",
                "kind": "hnsw",
                "hnswParameters": {
                    "m": 4,
                    "efConstruction": 400,
                    "efSearch": 500,
                    "metric": "cosine"
                }
            }
        ]
    }
}

To compile

• In ./Backend, add a nodemon.json file containing the following

{
    "env": {
        "SEARCH_API_ENDPOINT": "https://[instance].search.windows.net",
        "SEARCH_API_KEY": "[api-key]",
        "SEARCH_INDEX_NAME": "[index-name]",
        "AZURE_OPENAI_ENDPOINT": "https://[instance-name].azure.com/",
        "AZURE_OPENAI_API_KEY": "[api-key]",
        "AZURE_OPENAI_MODEL_EMBEDDINGS": "text-embedding-ada-002",
        "AZURE_OPENAI_MODEL_CHAT": "gpt-35-turbo-16k"
    }
}

• In ./Frontend, add a .env.development file containing the following

VITE_APP_NAME=DocChat
VITE_API_BASE_URL=http://127.0.0.1:3000

• From each directory (./Backend and ./Frontend), run the following commands to compile and run
  • pnpm install
  • pnpm run dev

Thoughts for improvement

• The app contains no authentication mechanism, this should be changed if being moved into production
• The app does not retain chat sessions between browser refreshes, this should be changed by storing chat conversation in a database (or by using a browser with IndexedDB)