Upload slide_generator.py with huggingface_hub

slide_generator.py

@@ -3,13 +3,11 @@ import json
 import re
 from huggingface_hub import InferenceClient
 
-# Model: Qwen/Qwen2.5-7B-Instruct — 7B parameters (total used: 7B ≤ 32B limit)
 MODEL_NAME = "Qwen/Qwen2.5-7B-Instruct"
 
 _api_key = os.environ.get("GROQ_API_KEY") or os.environ.get("HF_TOKEN")
 client = InferenceClient(api_key=_api_key)
 
-
 SYSTEM_PROMPT = (
     "You are an expert presentation designer. "
     "Return only valid JSON — no markdown fences, no extra text."
@@ -17,34 +15,33 @@ SYSTEM_PROMPT = (
 
 
 def _extract_json(text: str) -> dict:
-    """Robustly pull a JSON object out of the model response."""
-    # Try direct parse first
     try:
         return json.loads(text.strip())
     except json.JSONDecodeError:
         pass
-    # Strip markdown fences if present
     cleaned = re.sub(r"^```[a-z]*\n?", "", text.strip(), flags=re.MULTILINE)
     cleaned = re.sub(r"\n?```$", "", cleaned.strip(), flags=re.MULTILINE)
     try:
         return json.loads(cleaned.strip())
     except json.JSONDecodeError:
         pass
-    # Find first { ... } block
     match = re.search(r"\{.*\}", text, re.DOTALL)
     if match:
         return json.loads(match.group())
     raise ValueError("Could not extract JSON from model response.")
 
 
-def generate_presentation(topic: str, style: str, num_slides: int, audience: str, key_points: str) -> dict:
-    key_points_section = f"\nKey points to include: {key_points}" if key_points.strip() else ""
+def generate_presentation(topic: str, style: str, num_slides: int,
+                          audience: str, key_points: str) -> dict:
+    key_points_section = (
+        f"\nKey points to include: {key_points}" if key_points.strip() else ""
+    )
 
     user_prompt = f"""Create a {style.lower()} presentation about: "{topic}"
 Target audience: {audience}
 Number of slides: {num_slides} (including title slide){key_points_section}
 
-Return a JSON object with this exact structure:
+Return a JSON object with this EXACT structure:
 {{
   "title": "Main presentation title",
   "subtitle": "A compelling subtitle",
@@ -54,6 +51,7 @@ Return a JSON object with this exact structure:
       "type": "title",
       "title": "Presentation Title",
       "subtitle": "Subtitle or tagline",
+      "image_keyword": "relevant photo keyword for background",
       "speaker_notes": "Opening notes for the presenter"
     }},
     {{
@@ -61,6 +59,7 @@ Return a JSON object with this exact structure:
       "type": "content",
       "title": "Slide Title",
       "bullets": ["Point one", "Point two", "Point three"],
+      "image_keyword": "2-3 word photo search term relevant to this slide",
       "speaker_notes": "What to say during this slide"
     }}
   ]
@@ -69,6 +68,7 @@ Return a JSON object with this exact structure:
 Rules:
 - First slide must be type "title" with subtitle field
 - All other slides type "content" with bullets array (3-5 items, max 12 words each)
+- image_keyword: 2-3 words describing a real-world photo relevant to the slide (e.g. "solar panels field", "team meeting office", "brain neuroscience")
 - Speaker notes: 2-3 sentences
 - Total slides: exactly {num_slides}
 - Return only the JSON object, nothing else