Ensure cache is writtable

Dockerfile

@@ -1,31 +1,32 @@
 FROM python:3.11-slim
 
-# Set environment variables (may need to change authority)
+# Set environment variables
 ENV PYTHONDONTWRITEBYTECODE=1 \
-    PYTHONUNBUFFERED=1
+    PYTHONUNBUFFERED=1 \
+    TRANSFORMERS_CACHE=/app/model_cache \
+    HF_HOME=/app/.cache/huggingface \
+    SENTENCE_TRANSFORMERS_HOME=/app/.cache/huggingface/sentence-transformers
 
-# Create working directory (and authorised token)
+# Create working directory
 WORKDIR /app
 
-# Add to Dockerfile
-ENV TRANSFORMERS_CACHE=/app/model_cache
-ENV HF_HOME=/app/.cache/huggingface
-ENV SENTENCE_TRANSFORMERS_HOME=/app/.cache/huggingface/sentence-transformers
-
-# Install SSL root certs and system deps required by pymongo + DNS
+# Install system dependencies
 RUN apt-get update && apt-get install -y --no-install-recommends \
     ca-certificates curl dnsutils gcc openssl && \
     rm -rf /var/lib/apt/lists/*
 
-# Copy and install dependencies
+# Install Python dependencies
 COPY requirements.txt .
 RUN pip install --upgrade pip && pip install --no-cache-dir -r requirements.txt
 
-# Copy source code (app files)
+# Create writable model cache directory to avoid permission error
+RUN mkdir -p /app/model_cache /app/.cache/huggingface
+
+# Copy app source code
 COPY . .
 
-# Expose the port used by Uvicorn
+# Expose app port
 EXPOSE 7860
 
-# Run the FastAPI application using Uvicorn
+# Run app
 CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860", "--log-level", "debug"]

app.py

@@ -115,6 +115,7 @@ check_system_resources()
 @app.get("/")
 async def root() -> FileResponse:
     """Serve the single‑page app."""
+    logger.info("[STATIC] Serving frontend")
     return FileResponse(Path("statics/index.html"))
 
 
@@ -136,6 +137,7 @@ async def voice_transcribe(file: UploadFile = File(...)):  # noqa: B008
         question = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
         if not question:
             raise ValueError("Empty transcription")
+        logger.info(f"[VOICE] Detected transcribe: {question}")
         # ── 2. LLM answer
         prompt = build_prompt(question)
         # Gemini Flash 2.5 – tuned for short latency
@@ -145,6 +147,7 @@ async def voice_transcribe(file: UploadFile = File(...)):  # noqa: B008
                 contents=prompt
             )
         answer = response.text.strip()
+        logger.info(f"[LLM] Decision answer: {answer}")
         return JSONResponse(
             {
                 "question": question,