feat: implement AI agent layer with LangChain integration

- Create InterviewerAgent with Socratic questioning and RAG context
  - Build SynthesizerAgent for transcript segmentation and link generation
  - Integrate Google Gemini models (Flash for interviewing, Pro for synthesis)
  - Add structured output parsing for Zettel extraction and linking
  - Implement session termination detection with [END_SESSION] token
  - Add conversation context formatting and similarity-based neighbor filtering
  - Add vector service tests with mocked ChromaDB and embeddings
  - Test interviewer agent RAG conversations and session termination
  - Test synthesizer agent transcript formatting and neighbor analysis
  - Add prompt loader tests for external prompt system
  - Test all repository CRUD operations and database transactions

tests/test_vector_service.py

@@ -0,0 +1,121 @@
+"""Test vector service functionality."""
+import pytest
+import uuid
+from datetime import datetime
+from unittest.mock import AsyncMock, MagicMock, patch
+
+from app.services.vector import VectorService
+from app.data.models import Note
+
+
+class TestVectorService:
+    """Test VectorService operations."""
+    
+    @pytest.fixture
+    def sample_notes(self):
+        """Create sample notes for testing."""
+        session_id = uuid.uuid4()
+        return [
+            Note(
+                title="AI Ethics",
+                content="Artificial intelligence systems must be designed with ethical considerations in mind.",
+                tags=["ai", "ethics", "technology"],
+                session_id=session_id
+            ),
+            Note(
+                title="Machine Learning Bias", 
+                content="Bias in machine learning models can perpetuate unfair discrimination.",
+                tags=["ml", "bias", "fairness"],
+                session_id=session_id
+            )
+        ]
+    
+    @pytest.mark.asyncio
+    async def test_vector_service_initialization(self):
+        """Test that VectorService initializes correctly."""
+        with patch('app.services.vector.GoogleGenerativeAIEmbeddings'):
+            with patch('app.services.vector.chromadb.PersistentClient'):
+                service = VectorService()
+                assert service is not None
+    
+    @pytest.mark.asyncio 
+    async def test_add_notes_empty_list(self):
+        """Test adding empty list of notes."""
+        with patch('app.services.vector.GoogleGenerativeAIEmbeddings'):
+            with patch('app.services.vector.chromadb.PersistentClient'):
+                service = VectorService()
+                await service.add_notes([])  # Should not raise an error
+    
+    @pytest.mark.asyncio
+    async def test_add_notes_with_mocked_embeddings(self, sample_notes):
+        """Test adding notes with mocked embedding service."""
+        mock_embeddings = AsyncMock()
+        mock_embeddings.aembed_documents = AsyncMock(return_value=[[0.1, 0.2, 0.3], [0.4, 0.5, 0.6]])
+        
+        mock_collection = MagicMock()
+        mock_client = MagicMock()
+        mock_client.get_or_create_collection.return_value = mock_collection
+        
+        with patch('app.services.vector.GoogleGenerativeAIEmbeddings', return_value=mock_embeddings):
+            with patch('app.services.vector.chromadb.PersistentClient', return_value=mock_client):
+                service = VectorService()
+                await service.add_notes(sample_notes)
+                
+                # Verify embeddings were called
+                mock_embeddings.aembed_documents.assert_called_once()
+                # Verify collection.add was called
+                mock_collection.add.assert_called_once()
+    
+    @pytest.mark.asyncio
+    async def test_semantic_search_empty_collection(self):
+        """Test semantic search on empty collection."""
+        mock_collection = MagicMock()
+        mock_collection.count.return_value = 0
+        
+        mock_client = MagicMock()
+        mock_client.get_or_create_collection.return_value = mock_collection
+        
+        with patch('app.services.vector.GoogleGenerativeAIEmbeddings'):
+            with patch('app.services.vector.chromadb.PersistentClient', return_value=mock_client):
+                service = VectorService()
+                results = await service.semantic_search("test query")
+                assert results == []
+    
+    @pytest.mark.asyncio
+    async def test_semantic_search_with_results(self):
+        """Test semantic search with mocked results."""
+        mock_embeddings = AsyncMock()
+        mock_embeddings.aembed_query = AsyncMock(return_value=[0.1, 0.2, 0.3])
+        
+        mock_collection = MagicMock()
+        mock_collection.count.return_value = 2
+        mock_collection.query.return_value = {
+            "documents": [["AI ethics content", "ML bias content"]], 
+            "metadatas": [[{"title": "AI Ethics", "tags": "ai,ethics"}, {"title": "ML Bias", "tags": "ml,bias"}]],
+            "distances": [[0.2, 0.4]]
+        }
+        
+        mock_client = MagicMock()
+        mock_client.get_or_create_collection.return_value = mock_collection
+        
+        with patch('app.services.vector.GoogleGenerativeAIEmbeddings', return_value=mock_embeddings):
+            with patch('app.services.vector.chromadb.PersistentClient', return_value=mock_client):
+                service = VectorService()
+                results = await service.semantic_search("ethics")
+                
+                assert len(results) == 2
+                assert results[0].page_content == "AI ethics content"
+                assert results[0].metadata["title"] == "AI Ethics"
+                assert results[0].metadata["similarity_score"] == 0.8  # 1 - 0.2
+    
+    def test_reset_collection(self):
+        """Test collection reset functionality."""
+        mock_collection = MagicMock()
+        mock_client = MagicMock()
+        mock_client.get_or_create_collection.return_value = mock_collection
+        
+        with patch('app.services.vector.GoogleGenerativeAIEmbeddings'):
+            with patch('app.services.vector.chromadb.PersistentClient', return_value=mock_client):
+                service = VectorService()
+                service.reset_collection()  # Should not raise an error
+                mock_client.delete_collection.assert_called_once_with("skytalk_notes")