feat: implement NoteGenerator for LLM-based transcript summarization and add note management routes

src/api/notes_routes.py

@@ -281,8 +281,7 @@ async def process_video_task(
         from src.utils.model_loader import INFERENCE_MODE
 
         if INFERENCE_MODE == "groq":
-            from src.summarization.note_generator import parse_key_points_from_markdown
-            key_points_list = parse_key_points_from_markdown(summary_json.get("summary", ""))
+            key_points_list = summary_json.get("key_points", [])
         else:
             segments = summary_json.get("segments", [])
             key_points_list = [
@@ -298,13 +297,12 @@ async def process_video_task(
         )
 
         if INFERENCE_MODE == "groq":
-            from src.categorization.topic_classifier import classify_topic_groq
-            category = classify_topic_groq(resolved_video_title, summary_json.get("summary", ""))
+            category = summary_json.get("suggested_category", "Education")
             categories = [category]
         else:
             from src.categorization.topic_classifier import classify_topics
             raw_topics = summary_json.get("topics", [])
-            categories = classify_topics(raw_topics) if raw_topics else ["Education & Science"]
+            categories = classify_topics(raw_topics) if raw_topics else ["Education"]
 
         _set_task_status(task_id, "complete", "Generation completed successfully.")
         tasks[task_id]["notes"] = final_markdown

src/summarization/note_generator.py

@@ -406,6 +406,11 @@ class NoteGenerator:
 
         # 3. Detect language from the transcript (used by UI and metadata)
         detected_language = self._detect_language(transcript_text)
+        
+        # Override detected language based on the REDUCE step if available
+        llm_lang = overall_summary.get("language")
+        if llm_lang:
+            detected_language = "Arabic" if llm_lang == "ar" else "English"
 
         if INFERENCE_MODE == "groq":
             conclusion_text = (
@@ -433,11 +438,12 @@ class NoteGenerator:
         candidate = {
             "title": video_title,
             "detected_language": detected_language,
-            "summary": overall_summary,
+            "summary": overall_summary.get("markdown_summary", ""),
             "segments": segments_list,
-            "suggested_category": "General",
+            "suggested_category": overall_summary.get("category", "Education"),
             "conclusion": conclusion_text,
             "topics": topics_list,
+            "key_points": overall_summary.get("key_points", []),
         }
 
         # 3. VALIDATE — enforce the schema contract before returning