Merge pull request #37 from werther41/bugfix-outdated-model

update embedding model

Author: werther41

documents/QUICK_REFERENCE.md

@@ -15,8 +15,8 @@
 # Run tests
 ./scripts/test-all.sh
 
-# Trigger news ingestion
-curl http://localhost:3000/api/cron/retrieve-news
+# Trigger news ingestion (RSS fetch + text embedding generation)
+curl -X POST -H "Authorization: Bearer $CRON_SECRET" http://localhost:3000/api/cron/retrieve-news
 
 # Check topics
 curl http://localhost:3000/api/topics

documents/api-docs.md

@@ -606,13 +606,20 @@ Returns information about the most relevant news article without generating a fa
 
 ### 10. News Ingestion Cron Job
 
-**POST** `/api/cron/retrieve-news`
+**POST** `/api/cron/retrieve-news`  
+**GET** `/api/cron/retrieve-news` (same behavior; for manual/testing use)
 
-Fetches and stores news articles from RSS feeds. Requires authorization header.
+Fetches articles from configured RSS feeds (last 48 hours), generates text embeddings (Gemini `gemini-embedding-001`), and stores them in the database. Use this to manually trigger the same pipeline that runs on schedule.
 
 **Headers:**
 
-- `Authorization: Bearer {CRON_SECRET}`
+- `Authorization: Bearer {CRON_SECRET}` (required for both POST and GET unless `x-vercel-cron` is present)
+
+**Example (manual retrieval):**
+
+```bash
+curl -X POST -H "Authorization: Bearer $CRON_SECRET" http://localhost:3000/api/cron/retrieve-news
+```
 
 **Response:**
 
@@ -775,7 +782,7 @@ Uses Turso's native vector search with cosine similarity to find the most releva
 
 ### AI Integration
 
-- **Embedding Model**: Google Gemini `text-embedding-004` (768 dimensions)
+- **Embedding Model**: Google Gemini `gemini-embedding-001` (768 dimensions)
 - **LLM**: Google Gemini `gemini-2.0-flash-lite`
 - **Streaming**: Real-time fact generation with word-by-word display
 

lib/embeddings.ts

@@ -4,7 +4,7 @@ import { GoogleGenerativeAI } from "@google/generative-ai"
 const genAI = new GoogleGenerativeAI(process.env.GOOGLE_API_KEY!)
 
 /**
- * Generate embedding for text using Gemini text-embedding-004 model
+ * Generate embedding for text using Gemini gemini-embedding-001 model
  * @param text - The text to generate embedding for
  * @returns Promise<number[]> - Array of float32 values representing the embedding
  */
@@ -14,8 +14,11 @@ export async function generateEmbedding(text: string): Promise<number[]> {
       throw new Error("GOOGLE_API_KEY environment variable not set")
     }
 
-    const model = genAI.getGenerativeModel({ model: "text-embedding-004" })
-    const result = await model.embedContent(text)
+    const model = genAI.getGenerativeModel({ model: "gemini-embedding-001" })
+    const result = await model.embedContent({
+      content: { parts: [{ text }] },
+      output_dimensionality: 768,
+    } as unknown as Parameters<typeof model.embedContent>[0])
 
     return result.embedding.values
   } catch (error) {