Update UI and prompts

src/llm/processor.py

@@ -1,11 +1,13 @@
 import logging
 import os
+from posix import system
+from this import s
 from typing import Any, Dict, Optional
 
 from openai import OpenAI
 from pydantic import ValidationError
 
-from .models import ExtractionResult
+from .models import ExtractionResult, FilterResult
 from .prompts import EXTRACTION_SYSTEM_PROMPT, NOISE_FILTER_SYSTEM_PROMPT
 
 logger = logging.getLogger(__name__)
@@ -90,13 +92,28 @@ class LLMProcessor:
             logger.error(f"LLM Error: {e}")
             return ""
 
-    def filter_transcript(self, text: str, context: Optional[str] = None) -> str:
+    def filter_transcript(
+        self, text: str, context: Optional[str] = None
+    ) -> FilterResult:
         """
         Stage 1: Raw Transcript -> Filtered Text.
         """
-        result = self._call_llm(NOISE_FILTER_SYSTEM_PROMPT, text, context=context)
+        result = self._call_llm(
+            NOISE_FILTER_SYSTEM_PROMPT,
+            text,
+            context=context,
+            response_format={"type": "json_object"},
+        )
         logger.info(f"LLM Processor (Filter): {text} -> {result}")
-        return result
+
+        import json
+
+        try:
+            data = json.loads(result)
+            return FilterResult(**data)
+        except (json.JSONDecodeError, ValidationError) as e:
+            logger.error(f"Filter Parsing Error: {e}")
+            return FilterResult(contextual_info="", filtered_text=result)
 
     def extract_structured_data(
         self, filtered_text: str, context: Optional[str] = None
@@ -109,18 +126,18 @@ class LLMProcessor:
             # Using standard chat.completions.create with JSON mode for better compatibility with vLLM
             logger.info("LLM Processor (Extract): Sending request to backend...")
 
-            messages = [
-                {"role": "system", "content": EXTRACTION_SYSTEM_PROMPT},
-            ]
+            system_prompt = EXTRACTION_SYSTEM_PROMPT
             if context:
-                messages.append(
-                    {
-                        "role": "system",
-                        "content": f"Context from previous conversation:\n{context}",
-                    }
-                )
+                system_prompt += f"\n{context}"
+
+            messages = [
+                {"role": "system", "content": system_prompt},
+            ]
             messages.append({"role": "user", "content": filtered_text})
 
+            for message in messages:
+                logger.info(f"LLM Processor (Extract): Message: {message}")
+
             response = self.client.chat.completions.create(
                 model=self.model,
                 messages=messages,
@@ -142,15 +159,3 @@ class LLMProcessor:
             logger.error(f"Extraction Error: {e}")
             # Return an empty ExtractionResult if parsing fails
             return ExtractionResult()
-
-    def process_pipeline(
-        self, raw_text: str, context: Optional[str] = None
-    ) -> ExtractionResult:
-        """
-        Executes the two-stage pipeline: Raw Transcript -> Filtered Text -> Structured Data.
-        """
-        filtered_text = self.filter_transcript(raw_text, context=context)
-        if not filtered_text:
-            return ExtractionResult()
-
-        return self.extract_structured_data(filtered_text, context=context)