logs and unit tests

README.md

@@ -1,8 +1,8 @@
 ---
 title: Context-aware NLP classification platform with MCP
 emoji: 🧠
-colorFrom: red
-colorTo: deep red
+colorFrom: indigo
+colorTo: red
 sdk: docker
 app_file: Dockerfile
 pinned: false

app/classification/decision.py

@@ -2,7 +2,12 @@ from typing import Any
 from dataclasses import dataclass
 from app.config import get_settings
 
+from app.logging.inference_log import log_inference
+from app.logging.context_log import log_context_resolution  # ← added
+import logging
+
 settings = get_settings()
+logger = logging.getLogger(__name__)
 
 
 @dataclass
@@ -18,6 +23,7 @@ def classify_document(text: str, context: Any) -> ClassificationDecision:
     - Model prediction
     - Confidence threshold
     - Abstention logic
+    - Logs inference and context usage
     """
 
     from app.classification.model import Classifier
@@ -30,8 +36,9 @@ def classify_document(text: str, context: Any) -> ClassificationDecision:
     else:
         context_dict = {}
 
-    result = classifier.predict(text=text, context=context_dict)
+    logger.info("Classification request received", extra={"text": text[:100]})
 
+    result = classifier.predict(text=text, context=context_dict)
 
     label = result.get("label")
     confidence = result.get("confidence", 0.0)
@@ -40,6 +47,21 @@ def classify_document(text: str, context: Any) -> ClassificationDecision:
     if settings.ENABLE_ABSTENTION and confidence < settings.CONFIDENCE_THRESHOLD:
         label = None
         abstained = True
+        logger.warning(
+            "Low confidence classification, abstaining", extra={"confidence": confidence}
+        )
+
+    # Log to persistent JSON files
+    log_inference(
+        label=label,
+        confidence=confidence,
+        abstained=abstained,
+        text=text,
+        context=context_dict,
+    )
+
+    # Minimal addition: log context usage
+    log_context_resolution(context=context_dict, text=text)
 
     return ClassificationDecision(
         label=label,

app/context/resolver.py

@@ -0,0 +1,23 @@
+from app.logging.context_log import log_context_resolution
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+def resolve_context(text: str, metadata: dict = None) -> dict:
+    """
+    Placeholder for context resolution logic.
+    Logs which context sources were used via JSON audit.
+    """
+    # Dummy context (replace with actual MCP context resolution)
+    context = {
+        "summary": f"Context for '{text[:50]}'",
+        "sources": ["mcp_server_1", "mcp_server_2"],
+    }
+
+    logger.info("Context resolution performed", extra={"text": text[:100]})
+
+    # Log JSON audit
+    log_context_resolution(context, text, metadata)
+
+    return context

app/logging_config.py

@@ -0,0 +1,35 @@
+import logging
+import sys
+from pathlib import Path
+
+LOG_LEVEL = logging.INFO
+LOG_FORMAT = "%(asctime)s | %(levelname)s | %(name)s | %(message)s"
+
+def setup_logging():
+    handlers = []
+
+    # Always log to stdout (HF Spaces, Docker, CI)
+    stdout_handler = logging.StreamHandler(sys.stdout)
+    stdout_handler.setFormatter(logging.Formatter(LOG_FORMAT))
+    handlers.append(stdout_handler)
+
+    # Optional file logging (local only)
+    logs_dir = Path("logs")
+    try:
+        logs_dir.mkdir(exist_ok=True)
+        file_handler = logging.FileHandler(logs_dir / "app.log")
+        file_handler.setFormatter(logging.Formatter(LOG_FORMAT))
+        handlers.append(file_handler)
+    except Exception:
+        # Fail silently — file logs are non-critical
+        pass
+
+    logging.basicConfig(
+        level=LOG_LEVEL,
+        handlers=handlers,
+        force=True,  # overrides uvicorn defaults
+    )
+
+    # Reduce noise from dependencies
+    logging.getLogger("uvicorn.access").setLevel(logging.WARNING)
+    logging.getLogger("httpx").setLevel(logging.WARNING)

app/main.py

@@ -1,30 +1,50 @@
 from fastapi import FastAPI
 from fastapi.middleware.cors import CORSMiddleware
+from contextlib import asynccontextmanager
 
 from app.config import get_settings
 from app.api.routes import router as api_router
 
 # MCP (embedded mode)
-from app.orchestration.mcp_client import start_embedded_mcp_servers
-from app.orchestration.mcp_client import stop_embedded_mcp_servers
+from app.orchestration.mcp_client import start_embedded_mcp_servers, stop_embedded_mcp_servers
+
+from app.logging_config import setup_logging
+import logging
 
 
 def create_app() -> FastAPI:
     """
-    Application factory.
-
-    Ensures:
-    - Deterministic startup
-    - Clean separation of concerns
-    - HF Spaces compatibility
+    Application factory with lifespan handler for startup/shutdown.
     """
-
     settings = get_settings()
+    
+    setup_logging()
+    logger = logging.getLogger(__name__)
+
+    @asynccontextmanager
+    async def lifespan(app: FastAPI):
+        """Handles startup and shutdown with lifespan."""
+        # Startup
+        logger.info("Application startup initiated")
+        if settings.MCP_EMBEDDED:
+            logger.info("Starting embedded MCP servers")
+            start_embedded_mcp_servers()
+        logger.info("Application startup complete")
+
+        yield  # Control passes to FastAPI for request handling
+
+        # Shutdown
+        logger.info("Application shutdown initiated")
+        if settings.MCP_EMBEDDED:
+            logger.info("Stopping embedded MCP servers")
+            stop_embedded_mcp_servers()
+        logger.info("Application shutdown complete")
 
     app = FastAPI(
         title=settings.APP_NAME,
         debug=settings.DEBUG,
         version="0.1.0",
+        lifespan=lifespan,  # ← attach lifespan handler
     )
 
     # -------------------------
@@ -43,33 +63,8 @@ def create_app() -> FastAPI:
     # -------------------------
     app.include_router(api_router)
 
-    # -------------------------
-    # Lifecycle Events
-    # -------------------------
-    @app.on_event("startup")
-    async def on_startup() -> None:
-        """
-        Startup logic.
-
-        In HF Spaces:
-        - MCP servers are started embedded
-
-        Locally:
-        - MCP servers are external (docker-compose)
-        """
-        if settings.MCP_EMBEDDED:
-            start_embedded_mcp_servers()
-
-    @app.on_event("shutdown")
-    async def on_shutdown() -> None:
-        """
-        Graceful shutdown.
-        """
-        if settings.MCP_EMBEDDED:
-            stop_embedded_mcp_servers()
-
     return app
 
 
-# FastAPI entrypoint (required by HF Spaces and uvicorn)
+# FastAPI entrypoint
 app = create_app()

docs/TECH_DEBT.md

@@ -1,7 +1,7 @@
 ## Pydantic / FastAPI deprecations
 
-- Migrate BaseSettings Config → ConfigDict
-- Replace Field(example=...) with json_schema_extra
-- Replace @app.on_event with lifespan handler
+- Migrate BaseSettings `Config` → `ConfigDict`
+- Replace `Field(example=...)` with `json_schema_extra`
+- `datetime.utcnow()` usage in logging → switch to timezone-aware `datetime.now(datetime.UTC)`
 
 Status: deferred (non-blocking)

tests/conftest.py

@@ -0,0 +1,18 @@
+import sys
+from pathlib import Path
+
+# Ensure root folder is in Python path so `import app` works
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+import pytest
+from app.orchestration.mcp_client import start_embedded_mcp_servers, stop_embedded_mcp_servers
+
+@pytest.fixture(scope="session", autouse=True)
+def embedded_mcp_servers():
+    """
+    Starts embedded MCP servers for all tests that require MCP context.
+    Runs once per test session.
+    """
+    start_embedded_mcp_servers()
+    yield
+    stop_embedded_mcp_servers()

tests/test_classification.py

@@ -1,9 +1,16 @@
 from app.classification.decision import classify_document
+from app.orchestration.context_resolver import resolve_context
+
+def test_invoice_classification(embedded_mcp_servers):
+    # Ensure MCP context is loaded
+    context = resolve_context(
+        text="Invoice for Q4 2025 total amount $4,500",
+        metadata={"department": "finance"}
+    )
 
-def test_invoice_classification():
     result = classify_document(
         text="Invoice for Q4 2025 total amount $4,500",
-        context={}
+        context=context
     )
 
     assert result.label == "finance.invoice"
@@ -11,10 +18,15 @@ def test_invoice_classification():
     assert result.abstained is False
 
 
-def test_abstention_logic():
+def test_abstention_logic(embedded_mcp_servers):
+    context = resolve_context(
+        text="Random unrelated text with no meaning",
+        metadata={}
+    )
+
     result = classify_document(
         text="Random unrelated text with no meaning",
-        context={}
+        context=context
     )
 
     assert "confidence" in result.__dict__

tests/test_context_resolution.py

@@ -1,6 +1,6 @@
 from app.orchestration.context_resolver import resolve_context
 
-def test_context_resolution_embedded():
+def test_context_resolution_embedded(embedded_mcp_servers):
     context = resolve_context(
         text="Invoice for Q4 2025",
         metadata={"department": "finance"}

tests/test_mcp_servers.py

@@ -1,7 +1,7 @@
 from app.orchestration.context_resolver import resolve_context
 from app.config import get_settings
 
-def test_mcp_fail_fast_disabled():
+def test_mcp_fail_fast_disabled(embedded_mcp_servers):
     settings = get_settings()
     settings.MCP_FAIL_FAST = False
 

ui/streamlit_app.py

@@ -1,6 +1,18 @@
+import sys
+from pathlib import Path
+
+# -------------------------
+# Ensure 'app' folder is importable
+# -------------------------
+PROJECT_ROOT = Path(__file__).resolve().parent.parent
+if str(PROJECT_ROOT) not in sys.path:
+    sys.path.append(str(PROJECT_ROOT))
+
+# -------------------------
+# Imports
+# -------------------------
 import streamlit as st
 import requests
-from pathlib import Path
 from app.config import get_settings
 
 settings = get_settings()