add generating notes method

main.py

@@ -183,4 +183,52 @@ def transcribe_audio(audio_bytes: bytes) -> str:
     except Exception as e:
         print(f"ERROR: Hugging Face ASR API call failed: {e}")
         # Check for specific HF error types if possible
-        raise HTTPException(status_code=503, detail=f"Transcription service failed: {e}") # 503 Service Unavailable
+        raise HTTPException(status_code=503, detail=f"Transcription service failed: {e}") # 503 Service Unavailable
+
+
+def generate_notes_from_transcript(transcript: str) -> str:
+    """
+    Sends the transcript to the Hugging Face text generation LLM API.
+    """
+    if not hf_inference:
+        raise HTTPException(status_code=503, detail="Note generation service client not initialized.")
+    if not transcript:
+        print("Warning: Skipping note generation for empty transcript.")
+        return "Could not generate notes: Transcription was empty."
+
+    print(f"Generating notes for transcript (length {len(transcript)}) using {LLM_MODEL}...")
+
+    # --- Prompt Engineering: Crucial for good results! ---
+    # Be explicit about the desired output format and role.
+    prompt = f"""You are an expert note-taking assistant specializing in extracting key information from video transcripts. 
+    Please analyze the following transcript and generate concise, well-structured notes. 
+    Focus on the main topics, key points, important examples, definitions, and any conclusions presented. Use bullet points or numbered lists for clarity.
+
+    Transcript:
+    \"\"\"
+    {transcript}
+    \"\"\"
+
+    Structured Notes:"""
+
+    try:
+        # Use the textGeneration task for instruction-following models like Mistral
+        response = hf_inference.text_generation(
+            prompt=prompt,
+            model=LLM_MODEL,
+            max_new_tokens=1024,  # Max length of the *generated* notes. Adjust as needed.
+                                 # Longer videos might need more tokens for comprehensive notes.
+            temperature=0.7,     # Controls randomness (lower = more focused, higher = more creative)
+            repetition_penalty=1.1, # Slightly discourage repeating the same phrases
+            # Other parameters like top_p, top_k can also be tuned
+        )
+
+        # The response is usually the generated text directly for text-generation
+        # Sometimes it might include the prompt, so basic cleaning can help.
+        notes = response.strip()
+        print("Note generation successful.")
+        return notes
+
+    except Exception as e:
+        print(f"ERROR: Hugging Face LLM API call failed: {e}")
+        raise HTTPException(status_code=503, detail=f"Note generation service failed: {e}")