chore: update models to gemini-2.5/3.0 and add diagnostic scripts

Author: AKdaAKA

backend/list-models.js

@@ -0,0 +1,32 @@
+const { GoogleGenerativeAI } = require("@google/generative-ai");
+require('dotenv').config({ path: __dirname + '/.env' });
+
+async function listModels() {
+  const apiKey = process.env.GEMINI_API_KEY;
+  if (!apiKey) {
+    console.error("❌ GEMINI_API_KEY not found in backend/.env");
+    return;
+  }
+
+  try {
+    const response = await fetch(`https://generativelanguage.googleapis.com/v1beta/models?key=${apiKey}`);
+    const data = await response.json();
+    
+    if (data.error) {
+      console.error("❌ API Error:", data.error.message);
+      return;
+    }
+
+    console.log("\n=== Available Models ===");
+    data.models.forEach(m => {
+      const name = m.name.replace('models/', '');
+      const methods = m.supportedGenerationMethods.join(', ');
+      console.log(`- ${name.padEnd(25)} | Methods: ${methods}`);
+    });
+    console.log("========================\n");
+  } catch (error) {
+    console.error("❌ Failed to list models:", error.message);
+  }
+}
+
+listModels();

backend/routes/tutorRoutes.js

@@ -231,16 +231,6 @@ router.post('/chat', async (req, res, next) => {
     // Combine targeted and broad context (Targeted first)
     let context = [...finalTargeted, ...finalBroad].join('\n--- NEXT CHUNK ---\n');
 
-    // --- ENHANCEMENT: Inject Structured Syllabus Data (Database-driven) ---
-    try {
-      const syllabusContext = await syllabusService.getSyllabusContext(classCode);
-      if (syllabusContext) {
-        context = syllabusContext + "\n" + context;
-        console.log(`[Tutor] 📋 Injected structured syllabus context for ${classCode}`);
-      }
-    } catch (syllErr) {
-      console.warn(`[Tutor] ⚠️ Syllabus context injection failed:`, syllErr.message);
-    }
 
     if (!context || context.trim().length === 0) {
       context = "No specific course documents were found matching this query. Answer using your general knowledge as a Socratic tutor.";

backend/services/syllabusService.js

@@ -107,7 +107,7 @@ Return ONLY JSON matching this schema:
 
   console.log("[Syllabus] 🤖 Sending direct PDF to Native Google SDK...");
 
-  const modelsToTry = ["gemini-2.5-flash", "gemini-2.5-pro"];
+  const modelsToTry = ["gemini-2.5-flash", "gemini-3-flash-preview", "gemini-2.5-pro"];
   let lastError;
 
   for (const modelId of modelsToTry) {

backend/services/tutorService.js

@@ -26,14 +26,14 @@ function getRobustLLM(temperature = 0.2, maxRetries = 1) {
   });
 
   const fallbackLite = new ChatGoogleGenerativeAI({
-    model: "gemini-3.0-flash",
+    model: "gemini-3-flash-preview",
     apiKey: process.env.GEMINI_API_KEY,
     temperature,
     maxRetries: 1,
   });
 
   return primary.withFallbacks({
-    fallbacks: [fallbackPro]
+    fallbacks: [fallbackLite, fallbackPro]
   });
 }
 
@@ -57,8 +57,15 @@ function getFastLLM() {
     maxRetries: 1,
   });
 
+  const fallbackLite = new ChatGoogleGenerativeAI({
+    model: "gemini-3-flash-preview",
+    apiKey: process.env.GEMINI_API_KEY,
+    temperature: 0.1,
+    maxRetries: 1,
+  });
+
   return primary.withFallbacks({
-    fallbacks: [fallbackPro]
+    fallbacks: [fallbackLite, fallbackPro]
   });
 }
 

backend/services/vectorService.js

@@ -41,18 +41,32 @@ const courseData = [
 
 class GeminiEmbeddings {
   constructor() {
-    this.model = new GoogleGenerativeAIEmbeddings({
+    this.primary = new GoogleGenerativeAIEmbeddings({
       apiKey: process.env.GEMINI_API_KEY,
-      modelName: "gemini-embedding-002",
+      modelName: "gemini-embedding-001",
+    });
+    this.fallback = new GoogleGenerativeAIEmbeddings({
+      apiKey: process.env.GEMINI_API_KEY,
+      modelName: "gemini-embedding-2",
     });
   }
 
   async embedDocuments(documents) {
-    return this.model.embedDocuments(documents);
+    try {
+      return await this.primary.embedDocuments(documents);
+    } catch (err) {
+      console.warn(`[VectorStore] Primary embedding failed, trying fallback:`, err.message);
+      return await this.fallback.embedDocuments(documents);
+    }
   }
 
   async embedQuery(document) {
-    return this.model.embedQuery(document);
+    try {
+      return await this.primary.embedQuery(document);
+    } catch (err) {
+      console.warn(`[VectorStore] Primary embedding failed, trying fallback:`, err.message);
+      return await this.fallback.embedQuery(document);
+    }
   }
 }
 
@@ -213,54 +227,19 @@ async function getVectorStore(namespace = '') {
 
   if (totalVectors === 0 && process.env.FORCE_RESEED !== 'true') {
     console.log(`[VectorStore] Namespace '${namespace || 'default'}' is empty — seeding course data...`);
-    // Embed with retry for Gemini rate limits
-    let vectors;
-    const maxEmbedRetries = 3;
-    for (let attempt = 1; attempt <= maxEmbedRetries; attempt++) {
-      try {
-        vectors = await embeddings.embedDocuments(courseData);
-        break;
-      } catch (embedErr) {
-        const isRetryable = embedErr.message && (
-          embedErr.message.includes("429") ||
-          embedErr.message.includes("RESOURCE_EXHAUSTED") ||
-          embedErr.message.includes("quota") ||
-          embedErr.message.includes("503")
-        );
-        if (isRetryable && attempt < maxEmbedRetries) {
-          const waitTime = attempt === 1 ? 5000 : 15000;
-          console.warn(`[VectorStore] ⚠️ Embed rate limited (attempt ${attempt}). Retrying in ${waitTime / 1000}s...`);
-          await new Promise(res => setTimeout(res, waitTime));
-        } else {
-          console.error(`[VectorStore] ❌ Embed failed after ${attempt} attempts:`, embedErr.message);
-          throw embedErr;
-        }
-      }
-    }
-
+    const vectors = await embeddings.embedDocuments(courseData);
     const records = courseData.map((text, i) => ({
       id: `course-${i}`,
       values: vectors[i],
       metadata: { text, source: 'course-data', chunk: i },
     }));
 
-    // Upsert with retry
+    // Upsert directly into the given namespace
     const targetIndex = namespace ? pineconeIndex.namespace(namespace) : pineconeIndex;
+
     const batchSize = 100;
     for (let i = 0; i < records.length; i += batchSize) {
-      let upserted = false;
-      for (let attempt = 1; attempt <= 3; attempt++) {
-        try {
-          await targetIndex.upsert({ records: records.slice(i, i + batchSize) });
-          upserted = true;
-          break;
-        } catch (upsertErr) {
-          const waitTime = attempt === 1 ? 3000 : 10000;
-          console.warn(`[VectorStore] ⚠️ Upsert batch ${i} failed (attempt ${attempt}). Retrying in ${waitTime / 1000}s...`);
-          await new Promise(res => setTimeout(res, waitTime));
-          if (attempt === 3) throw upsertErr;
-        }
-      }
+      await targetIndex.upsert({ records: records.slice(i, i + batchSize) });
     }
     console.log(`[VectorStore] Seeded ${courseData.length} vectors into Pinecone namespace: '${namespace || 'default'}'.`);
   } else {

backend/test-vector-service.js

@@ -0,0 +1,42 @@
+const { GeminiEmbeddings, getClassVectorStore } = require('./services/vectorService');
+require('dotenv').config({ path: __dirname + '/.env' });
+
+async function runTest() {
+  console.log("🚀 Starting Vector Service Test...");
+
+  // 1. Test Embeddings
+  const embeddings = new GeminiEmbeddings();
+  console.log("\n--- Step 1: Testing Embeddings ---");
+  try {
+    // Note: this will try primary then fallback
+    const queryVector = await embeddings.embedQuery("What is probability?");
+    console.log(`✅ Embedding successful! Vector length: ${queryVector.length}`);
+  } catch (err) {
+    console.error(`❌ Embedding failed: ${err.message}`);
+  }
+
+  // 2. Test Pinecone Connectivity
+  console.log("\n--- Step 2: Testing Pinecone Search ---");
+  const testClass = "CS3341"; // Example class
+  try {
+    const vectorStore = await getClassVectorStore(testClass);
+    console.log(`✅ VectorStore initialized for ${testClass}`);
+    
+    console.log(`Performing similarity search...`);
+    const results = await vectorStore.similaritySearch("probability", 1);
+    
+    if (results.length > 0) {
+      console.log(`✅ Search successful! Found ${results.length} result(s).`);
+      console.log(`Top result: "${results[0].pageContent.substring(0, 100)}..."`);
+    } else {
+      console.log(`ℹ️ Search executed but returned 0 results (namespace might be empty).`);
+    }
+  } catch (err) {
+    console.error(`❌ Pinecone test failed: ${err.message}`);
+  }
+
+  console.log("\n--- Test Complete ---");
+  process.exit(0);
+}
+
+runTest();