Merge pull request #1 from Tsukimarf/dependabot/github_actions/develop/actions/upload-artifact-5.0.0

Author: Tsukimarf

.github/copilot-instructions.md

@@ -0,0 +1,123 @@
+# Call Center AI - Developer Guide
+
+This is an AI-powered call center solution built with FastAPI, Azure Communication Services, and Azure OpenAI. The system handles real-time phone calls with intelligent conversation flow, claim processing, and multi-modal communication.
+
+## Architecture Overview
+
+The application follows a **modular, event-driven architecture** with clear separation between:
+
+- **Event handlers** (`app/helpers/call_events.py`) - Process Communication Services webhooks
+- **LLM orchestration** (`app/helpers/llm_worker.py`, `app/helpers/call_llm.py`) - Manage AI conversations
+- **Persistence layer** (`app/persistence/`) - Abstract interfaces for storage (Cosmos DB, Redis, AI Search)
+- **Models** (`app/models/`) - Pydantic models for data validation and serialization
+
+Key architectural patterns:
+
+- **Interface-based design**: All persistence layers implement abstract base classes (e.g., `ICache`, `IStore`, `ISearch`)
+- **Configuration-driven**: Single YAML config file drives all Azure resource connections
+- **Async-first**: Uses `asyncio`, `aiohttp`, and async Azure SDKs throughout
+- **Telemetry integration**: OpenTelemetry spans wrap all major operations
+
+## Core Data Flow
+
+1. **Inbound calls** → Azure Communication Services → Event Grid → FastAPI webhooks (`/webhook/*`)
+2. **Real-time audio** → WebSocket streaming → Speech-to-Text → LLM processing → Text-to-Speech
+3. **Conversation state** stored in `CallStateModel` → persisted to Cosmos DB
+4. **Claims data** extracted via LLM tools → structured in configurable schema
+
+## Development Workflow
+
+### Local Development
+
+```bash
+# Setup environment (uses uv for Python package management)
+make install
+
+# Run locally with auto-reload
+make dev
+
+# Sync remote Azure config to local
+make name=my-rg-name sync-local-config
+```
+
+### Configuration Management
+
+- **Local**: `config.yaml` (copy from `config-local-example.yaml`)
+- **Remote**: Azure App Configuration + environment variables
+- **Validation**: Pydantic models in `app/helpers/config_models/`
+
+### Testing Patterns
+
+- **Unit tests**: Mock Azure clients using custom mock classes (see `tests/conftest.py`)
+- **LLM evaluation**: Uses DeepEval for conversation quality assessment
+- **Async testing**: `pytest-asyncio` for async test functions
+
+## Key Components to Understand
+
+### Event-Driven Call Handling
+
+- `on_call_connected()` - Initiates conversation flow
+- `on_automation_recognize_error()` - Handles speech recognition failures with retry logic
+- `on_audio_connected()` - Starts real-time audio streaming for low-latency responses
+
+### LLM Integration
+
+- **Dual LLM strategy**: Fast model (`gpt-4.1-nano`) for real-time, slow model (`gpt-4.1`) for complex reasoning
+- **Tool calling**: LLM can execute predefined functions for claim updates, SMS sending, call transfers
+- **Context management**: Conversation history maintained in `CallStateModel.messages`
+
+### Audio Processing
+
+- **Noise reduction**: `noisereduce` library for audio cleanup
+- **Multi-language TTS**: Azure Cognitive Services with voice selection based on conversation language
+- **Streaming optimizations**: Audio chunks processed in real-time to minimize latency
+
+## Deployment & Infrastructure
+
+### Azure Resources (Bicep)
+
+- **Core**: `cicd/bicep/main.bicep` → `app.bicep` (modular deployment)
+- **Container deployment**: GitHub Actions builds container → Azure Container Apps
+- **Resource naming**: Uses sanitized instance names with consistent tagging
+
+### Makefile Commands
+
+```bash
+make deploy name=my-rg-name          # Deploy to Azure
+make logs name=my-rg-name            # View application logs
+make version                         # Get semantic version
+make tunnel-create                   # Setup dev tunnel for webhooks
+```
+
+## Project-Specific Conventions
+
+### Code Organization
+
+- **Helpers**: Pure functions and utilities, no state management
+- **Models**: Pydantic models with validation, used for API contracts and database schemas
+- **Persistence**: Interface implementations, each Azure service has its own module
+
+### Error Handling
+
+- **Graceful degradation**: Call continues even if non-critical services fail
+- **Retry logic**: Uses `tenacity` for resilient external service calls
+- **Structured logging**: `structlog` with OpenTelemetry correlation IDs
+
+### Configuration Validation
+
+All config is validated at startup using Pydantic. Missing or invalid Azure credentials cause immediate startup failure with clear error messages.
+
+## Integration Points
+
+- **Azure Communication Services**: Call automation, SMS, phone number management
+- **Azure OpenAI**: Dual deployment strategy (fast/slow models) with embedding search
+- **Azure AI Search**: RAG for training data and knowledge base
+- **Azure Cosmos DB**: Conversation persistence with partitioning by phone number
+- **Redis**: Caching layer for frequently accessed data and session state
+
+## Development Tips
+
+- **Mock Azure services** in tests using the patterns in `tests/conftest.py`
+- **Use async context managers** for resource cleanup (see `@asynccontextmanager` in `main.py`)
+- **Follow the interface pattern** when adding new persistence layers
+- **Test conversation flows** using the YAML-based conversation definitions in `tests/conversations.yaml`

.github/workflows/pipeline.yaml

@@ -145,7 +145,7 @@ jobs:
         run: make test-${{ matrix.step }} version_full=${{ needs.init.outputs.VERSION_FULL }}
 
       - name: Upload artifacts
-        uses: actions/upload-artifact@v4.6.0
+        uses: actions/upload-artifact@v5.0.0
         if: always()
         with:
           compression-level: 9