modify llm agent to compulsorily identify and scrape all provided fields

llm_agent.py

@@ -137,49 +137,30 @@ class LLMJobRefiner:
             return ""
 
     async def refine_job_data(self, raw_data: Dict[str, Any], target_field: str) -> Dict[str, Any]:
-        # Clean the raw HTML content for better LLM processing
         page_content = raw_data.get('page_content', '')
         cleaned_content = self._clean_html_for_llm(page_content)
-        
-        # Get job_id and url from raw data
         job_id = raw_data.get('job_id', 'unknown')
         url = raw_data.get('url', 'N/A')
         
         prompt = f"""
-        You are a job posting data extractor with two modes:
+        You are a job posting data extractor.
         
-        CRITICAL INSTRUCTIONS FOR TEXT FIELDS:
-        - description: Extract the COMPLETE job description text (all paragraphs, bullet points, everything)
-        - requirements: Extract the COMPLETE requirements section text if present (all details, don't summarize)
-        - qualifications: Extract the COMPLETE qualifications section text if present (all details, don't summarize)
-        - If these sections are not explicitly labeled but exist in the content, extract the relevant portions
+        EXTRACT EXACT TEXT - DO NOT SUMMARIZE, PARAPHRASE, OR INVENT.
         
-        PRIMARY MODE (PREFERRED):
-        - Extract EXACT text as it appears on the page for all fields
-        - DO NOT summarize, paraphrase, or interpret
-        - Copy verbatim content including original wording and formatting
+        For these critical fields, follow these rules:
+        - description: Extract ALL job description text. If ANY job details exist (duties, responsibilities, overview), include them. Only use "Not provided" if absolutely no description exists.
+        - requirements: Extract ALL requirements text. If ANY requirements exist (skills, experience, education needed), include them. Only use "Not provided" if none exist.
+        - qualifications: Extract ALL qualifications text. If ANY qualifications exist, include them. Only use "Not provided" if none exist.
         
+        REQUIRED FIELDS (must have valid values, never "N/A"):
+        - title, company_name, job_id, url
         
-        FALLBACK MODE (ONLY IF FIELD IS MISSING):
-        - If a field is NOT explicitly stated anywhere in the content, you MAY infer it using clear contextual clues
-        - Inference rules:
-          * company_name: Look for patterns like "at [Company]", "Join [Company]", "[Company] is hiring"
-          * nature_of_work: Look for "remote", "onsite", "hybrid", "work from home", "office-based"
-          * location: Extract city/state/country mentions near job title or details
-          * title: Use the largest/primary heading if no explicit "job title" label exists
-
-        REQUIRED FIELDS (must always have a value):
-        - title: Exact job title or best inference
-        - company_name: Exact company name or best inference  
-        - job_id: Use provided: {job_id}
-        - url: Use provided: {url}
-
-        OPTIONAL FIELDS (use exact text or "N/A" if not present and not inferable):
-        - salary_range
-        - nature_of_work
+        OPTIONAL FIELDS (can be "Not provided"):
+        - location, salary_range, nature_of_work
         
         Page Content:
         {cleaned_content}
+        
         Response format (ONLY return this JSON):
         {{
             "title": "...",
@@ -202,15 +183,32 @@ class LLMJobRefiner:
             )
             refined_data = self._parse_llm_response(response_text)
             
-            # Final validation - ensure required fields are present and meaningful
-            if refined_data:
-                required_fields = ['title', 'company_name', 'job_id', 'url']
-                for field in required_fields:
-                    if not refined_data.get(field) or refined_data[field] in ["N/A", "", "Unknown", "Company", "Job"]:
-                        return None  # LLM failed to extract properly
+            if not refined_data:
+                return None
                 
-                return refined_data
-            return None
+            # Validate required fields
+            required_fields = ['title', 'company_name', 'job_id', 'url']
+            for field in required_fields:
+                if not refined_data.get(field) or refined_data[field].strip() in ["N/A", "", "Unknown", "Company", "Job"]:
+                    return None
+            
+            # CRITICAL: Validate content fields - check if they SHOULD exist
+            content_fields = ['description', 'requirements', 'qualifications']
+            cleaned_original = cleaned_content.lower()
+            
+            # Simple heuristic: if page contains job-related keywords, content fields should NOT be "Not provided"
+            job_indicators = ['responsibilit', 'duties', 'require', 'qualifi', 'skill', 'experienc', 'educat', 'degree', 'bachelor', 'master']
+            has_job_content = any(indicator in cleaned_original for indicator in job_indicators)
+            
+            if has_job_content:
+                for field in content_fields:
+                    value = refined_data.get(field, "").strip()
+                    if value in ["Not provided", "N/A", ""]:
+                        # LLM failed to extract existing content
+                        print(f"  ⚠️ LLM returned '{value}' for {field} but job content appears present")
+                        return None
+            
+            return refined_data
             
         except Exception as e:
             print(f"LLM refinement failed: {str(e)}")