A/better rag (#2852)

* Enhance Azure OpenAI integration: update environment variables, improve client initialization with connectivity validation, and refactor embedder usage.

* Refactor code structure for improved readability and maintainability

* Refactor Azure OpenAI client initialization: streamline connection testing by using a dedicated utility function and remove redundant code.

* Add test utility for Azure OpenAI client: implement connection testing with timeout handling and model type support

* Update ingest.config.ts

* lodash-es

* fix ts

* restore system prompt

Author: TheSonOfThomp

apps/chatbot-server/.env.example

@@ -9,14 +9,15 @@ MONGODB_APP_NAME=LeafyGreenAI
 VECTOR_SEARCH_INDEX_NAME="vector_index" # or whatever your index name is
 MONGODB_DATABASE_NAME="mongodb-chatbot-framework-chatbot" # or whatever your database name is. must contain vector search index.
 
-# OpenAI config
-OPENAI_API_KEY=<OpenAI API key>
-OPENAI_EMBEDDING_MODEL="text-embedding-ada-002" # or other model
-OPENAI_CHAT_COMPLETION_MODEL="gpt-3.5-turbo" # or other model
-
 # Azure OpenAI config
 AZURE_OPENAI_API_KEY=<YOUR_AZURE_API_KEY1>
 AZURE_OPENAI_API_KEY=<YOUR_AZURE_API_KEY2>
 AZURE_OPENAI_ENDPOINT=https://<your-instance>.openai.azure.com/
+
 AZURE_OPENAI_EMBEDDING_MODEL=text-embedding-ada-002
-AZURE_OPENAI_CHAT_COMPLETION_MODEL=gpt-3.5-turbo
+AZURE_OPENAI_API_EMBEDDING_DEPLOYMENT_NAME=text-embedding-ada-002
+AZURE_OPENAI_API_EMBEDDING_DEPLOYMENT_URL=https://<your-instance>.openai.azure.com/openai/deployments/text-embedding-ada-002/embeddings?api-version=2023-05-15
+
+AZURE_OPENAI_CHAT_COMPLETION_MODEL=gpt-4.1
+AZURE_OPENAI_API_CHAT_COMPLETION_DEPLOYMENT_NAME=gpt-4.1
+AZURE_OPENAI_API_CHAT_COMPLETION_DEPLOYMENT_URL=https://<your-instance>.openai.azure.com/openai/deployments/gpt-4.1/chat/completions?api-version=2025-01-01-preview

apps/chatbot-server/package.json

@@ -1,9 +1,10 @@
 {
-  "name": "@leafygreen-ui/chatbot-server",
+  "name": "lg-chatbot-server",
   "version": "0.0.1",
   "description": "",
   "type": "module",
-  "main": "index.js",
+  "main": "dist/index.js",
+  "module": "dist/esm/index.js",
   "publishConfig": {
     "access": "restricted"
   },
@@ -22,12 +23,14 @@
     "dotenv": "^16.5.0",
     "express": "^4.18.2",
     "jsdom": "^26.1.0",
+    "lodash-es": "^4.17.21",
     "mongodb-chatbot-server": "^0.11.0",
     "mongodb-rag-core": "^0.7.0"
   },
   "devDependencies": {
     "mongodb-rag-ingest": "^0.3.1",
     "nodemon": "^3.0.1",
-    "tsx": "^4.19.4"
+    "tsx": "^4.19.4",
+    "typescript": "^5.8.0"
   }
 }

apps/chatbot-server/src/index.ts

@@ -10,8 +10,12 @@ import { initChatBot } from './init';
 // System prompt for chatbot
 const systemPrompt: SystemPrompt = {
   role: 'system',
-  content: `You are an assistant to users of the MongoDB Chatbot Framework.
+  content: `You are an assistant to engineers and product designers using the LeafyGreen design system.
 Answer their questions about the framework in a friendly conversational tone.
+
+For questions regarding engineering, and react components, provide code examples.
+For questions regarding design and UX guidelines, provide sources.
+
 Format your answers in Markdown.
 Be concise in your answers.
 `,

apps/chatbot-server/src/ingest/ingest.config.ts

@@ -8,14 +8,13 @@ import { loadEnvVars } from '../utils/loadEnv';
 import { makeEmbedder } from '../utils/makeEmbedder';
 
 import { leafygreenGithubSourceConstructor } from './sources/github-leafygreen-ui';
-import { mongoDbChatbotFrameworkDocsDataSourceConstructor } from './sources/github-mdb-chatbot-framework';
 import { webSourceConstructor } from './utils/webSourceConstructor';
 
 // Load project environment variables
 const {
   MONGODB_CONNECTION_URI,
   MONGODB_DATABASE_NAME,
-  AZURE_OPENAI_DEPLOYMENT,
+  AZURE_OPENAI_EMBEDDING_MODEL,
 } = loadEnvVars();
 
 export default {
@@ -25,7 +24,7 @@ export default {
       connectionUri: MONGODB_CONNECTION_URI,
       databaseName: MONGODB_DATABASE_NAME,
       searchIndex: {
-        embeddingName: AZURE_OPENAI_DEPLOYMENT,
+        embeddingName: AZURE_OPENAI_EMBEDDING_MODEL,
       },
     }),
   pageStore: () =>
@@ -39,6 +38,11 @@ export default {
       databaseName: MONGODB_DATABASE_NAME,
       entryId: 'all',
     }),
+  chunkOptions: () => ({
+    minChunkSize: 15,
+    maxChunkSize: 1000,
+    overlap: 100,
+  }),
   // Add data sources here
   dataSources: async () => {
     return Promise.all([
@@ -51,7 +55,6 @@ export default {
         'https://www.w3.org/WAI/standards-guidelines/wcag',
         'https://atomicdesign.bradfrost.com/table-of-contents',
       ].map(source => webSourceConstructor(source, {})),
-      mongoDbChatbotFrameworkDocsDataSourceConstructor(),
       leafygreenGithubSourceConstructor(),
     ]);
   },

apps/chatbot-server/src/ingest/utils/webSourceConstructor.ts

@@ -1,5 +1,6 @@
 /* eslint-disable no-console */
 import { recursiveCrawlFromBaseURL } from '@lg-tools/crawler';
+import { trimEnd } from 'lodash-es';
 import { Page } from 'mongodb-rag-core';
 import { type DataSource } from 'mongodb-rag-core/dataSources';
 
@@ -16,6 +17,7 @@ export async function webSourceConstructor(
   source: string,
   options?: WebSourceConstructorOptions,
 ): Promise<DataSource> {
+  source = trimEnd(source, '/');
   const { maxDepth = 3, verbose = false } = {
     maxDepth: 3,
     verbose: false,

apps/chatbot-server/src/init.ts

@@ -18,6 +18,7 @@ import { AzureOpenAI } from 'mongodb-rag-core/openai';
 
 import { loadEnvVars } from './utils/loadEnv';
 import { makeEmbedder } from './utils/makeEmbedder';
+import { testOpenAIClient } from './utils/testOpenAIClient';
 
 export async function initChatBot(): Promise<{
   llm: ChatLlm;
@@ -34,20 +35,27 @@ export async function initChatBot(): Promise<{
     AZURE_OPENAI_ENDPOINT,
     AZURE_OPENAI_API_KEY,
     AZURE_OPENAI_EMBEDDING_MODEL,
+    AZURE_OPENAI_API_CHAT_COMPLETION_DEPLOYMENT_NAME,
     AZURE_OPENAI_CHAT_COMPLETION_MODEL,
   } = loadEnvVars();
 
   const azureOpenAIChatClient = new AzureOpenAI({
     endpoint: AZURE_OPENAI_ENDPOINT,
     apiKey: AZURE_OPENAI_API_KEY,
     apiVersion: '2024-04-01-preview',
-    deployment: AZURE_OPENAI_CHAT_COMPLETION_MODEL,
+    deployment: AZURE_OPENAI_API_CHAT_COMPLETION_DEPLOYMENT_NAME,
+  });
+
+  testOpenAIClient({
+    client: azureOpenAIChatClient,
+    model: AZURE_OPENAI_CHAT_COMPLETION_MODEL,
+    type: 'chat',
   });
 
   // Chatbot LLM for responding to the user's query.
   const llm = makeOpenAiChatLlm({
     openAiClient: azureOpenAIChatClient,
-    deployment: AZURE_OPENAI_CHAT_COMPLETION_MODEL,
+    deployment: AZURE_OPENAI_API_CHAT_COMPLETION_DEPLOYMENT_NAME,
     openAiLmmConfigOptions: {
       temperature: 0.5,
     },
@@ -100,6 +108,7 @@ export async function initChatBot(): Promise<{
   
   
   User query: ${originalUserMessage}`;
+
     return { role: 'user', content: contentForLlm };
   };
 
@@ -114,6 +123,10 @@ export async function initChatBot(): Promise<{
   const mongodbClient = new MongoClient(MONGODB_CONNECTION_URI);
   const conversations = makeMongoDbConversationsService(
     mongodbClient.db(MONGODB_DATABASE_NAME),
+    {
+      LLM_NOT_WORKING: 'LLM_NOT_WORKING',
+      NO_RELEVANT_CONTENT: 'NO_RELEVANT_CONTENT',
+    },
   );
 
   return {

apps/chatbot-server/src/utils/loadEnv.ts

@@ -15,12 +15,14 @@ export function loadEnvVars() {
     MONGODB_APP_NAME,
     MONGODB_DATABASE_NAME,
     VECTOR_SEARCH_INDEX_NAME,
-    OPENAI_API_KEY,
-    OPENAI_EMBEDDING_MODEL,
     AZURE_OPENAI_ENDPOINT,
     AZURE_OPENAI_API_KEY,
     AZURE_OPENAI_EMBEDDING_MODEL,
+    AZURE_OPENAI_API_EMBEDDING_DEPLOYMENT_NAME,
+    AZURE_OPENAI_API_EMBEDDING_DEPLOYMENT_URL,
     AZURE_OPENAI_CHAT_COMPLETION_MODEL,
+    AZURE_OPENAI_API_CHAT_COMPLETION_DEPLOYMENT_NAME,
+    AZURE_OPENAI_API_CHAT_COMPLETION_DEPLOYMENT_URL,
   } = process.env;
   const requiredEnvVars = {
     MONGODB_USER,
@@ -29,12 +31,14 @@ export function loadEnvVars() {
     MONGODB_APP_NAME,
     MONGODB_DATABASE_NAME,
     VECTOR_SEARCH_INDEX_NAME,
-    OPENAI_API_KEY,
-    OPENAI_EMBEDDING_MODEL,
     AZURE_OPENAI_ENDPOINT,
     AZURE_OPENAI_API_KEY,
     AZURE_OPENAI_EMBEDDING_MODEL,
+    AZURE_OPENAI_API_EMBEDDING_DEPLOYMENT_NAME,
+    AZURE_OPENAI_API_EMBEDDING_DEPLOYMENT_URL,
     AZURE_OPENAI_CHAT_COMPLETION_MODEL,
+    AZURE_OPENAI_API_CHAT_COMPLETION_DEPLOYMENT_NAME,
+    AZURE_OPENAI_API_CHAT_COMPLETION_DEPLOYMENT_URL,
   } as const;
 
   for (const [name, value] of Object.entries(requiredEnvVars)) {
@@ -43,8 +47,10 @@ export function loadEnvVars() {
 
   const MONGODB_CONNECTION_URI = `mongodb+srv://${MONGODB_USER}:${MONGODB_PASSWORD}@${MONGODB_PROJECT_URL}/?retryWrites=true&w=majority&appName=${MONGODB_APP_NAME}`;
 
-  return { ...requiredEnvVars, MONGODB_CONNECTION_URI } as Record<
-    string,
-    string
-  >;
+  return {
+    ...requiredEnvVars,
+    MONGODB_CONNECTION_URI,
+  } as Record<keyof typeof requiredEnvVars, string> & {
+    MONGODB_CONNECTION_URI: string;
+  };
 }

apps/chatbot-server/src/utils/makeEmbedder.ts

@@ -9,13 +9,15 @@ import { makeEmbeddingClient } from './makeEmbeddingClient';
  * and in the chatbot server.
  */
 export const makeEmbedder = () => {
-  const { AZURE_OPENAI_DEPLOYMENT } = loadEnvVars();
+  const { AZURE_OPENAI_API_EMBEDDING_DEPLOYMENT_NAME } = loadEnvVars();
 
   const azureClient = makeEmbeddingClient();
 
-  return makeOpenAiEmbedder({
+  const embedder = makeOpenAiEmbedder({
     openAiClient: azureClient,
-    deployment: AZURE_OPENAI_DEPLOYMENT,
+    deployment: AZURE_OPENAI_API_EMBEDDING_DEPLOYMENT_NAME,
     backoffOptions: {},
   });
+
+  return embedder;
 };

apps/chatbot-server/src/utils/makeEmbeddingClient.ts

@@ -1,6 +1,7 @@
 import { AzureOpenAI } from 'mongodb-rag-core/openai';
 
 import { loadEnvVars } from './loadEnv';
+import { testOpenAIClient } from './testOpenAIClient';
 
 /**
  * Returns an consistent Azure OpenAI client
@@ -14,15 +15,30 @@ export const makeEmbeddingClient = () => {
   const {
     AZURE_OPENAI_API_KEY,
     AZURE_OPENAI_ENDPOINT,
-    AZURE_OPENAI_DEPLOYMENT,
+    AZURE_OPENAI_API_EMBEDDING_DEPLOYMENT_NAME,
   } = loadEnvVars();
 
   const azureClient = new AzureOpenAI({
     endpoint: AZURE_OPENAI_ENDPOINT,
     apiKey: AZURE_OPENAI_API_KEY,
     apiVersion: '2024-04-01-preview',
-    deployment: AZURE_OPENAI_DEPLOYMENT,
+    deployment: AZURE_OPENAI_API_EMBEDDING_DEPLOYMENT_NAME,
   });
 
+  // Validate the Azure client instance
+  if (!azureClient) {
+    throw new Error('Failed to initialize Azure OpenAI client.');
+  }
+
+  testOpenAIClient({
+    client: azureClient,
+    model: AZURE_OPENAI_API_EMBEDDING_DEPLOYMENT_NAME,
+    type: 'embedding',
+  });
+
+  // eslint-disable-next-line no-console
+  console.log('✅ Successfully connected to Azure OpenAI client');
+
   return azureClient;
+  // This is a minimal request to validate connectivity
 };

apps/chatbot-server/src/utils/testOpenAIClient.ts

@@ -0,0 +1,49 @@
+import { AzureOpenAI } from 'mongodb-rag-core/openai';
+
+interface TestOpenAIClientParams {
+  client: AzureOpenAI;
+  model: string;
+  type?: 'embedding' | 'chat';
+  timeoutMs?: number;
+}
+
+export const testOpenAIClient = async ({
+  client,
+  model,
+  type = 'embedding',
+  timeoutMs = 5000,
+}: TestOpenAIClientParams) => {
+  try {
+    // eslint-disable-next-line no-console
+    console.log(`✨ Testing Azure OpenAI client with ${type} model...`);
+
+    // Define the test request based on model type
+    const testPromise =
+      type === 'embedding'
+        ? client.embeddings.create({
+            input: 'test connection',
+            model,
+          })
+        : client.chat.completions.create({
+            messages: [{ role: 'user', content: 'Hello, test connection' }],
+            model,
+          });
+
+    // Set a timeout to catch if the request hangs
+    const timeoutPromise = new Promise((_, reject) => {
+      setTimeout(
+        () => reject(new Error('Azure OpenAI client connection timed out')),
+        timeoutMs,
+      );
+    });
+
+    // Race the request against the timeout
+    return await Promise.race([testPromise, timeoutPromise]);
+  } catch (error) {
+    throw new Error(
+      `Failed to connect to Azure OpenAI: ${
+        error instanceof Error ? error.message : String(error)
+      }`,
+    );
+  }
+};