Update app.py

app.py

@@ -29,22 +29,85 @@ Be friendly and conversational. Ask follow-up questions naturally. When appropri
 but respect their boundaries. Once you believe you have gathered sufficient information (or if the user indicates they
 have nothing more to share), let them know they can click 'Generate Profile' to proceed.
 """
+EXTRACTION_PROMPT = """You are a data extraction specialist. Your task is to:
+1. Read through the provided conversation
+2. Identify relevant information across 9 categories:
+   - Work History & Experience (jobs, roles, companies)
+   - Salary & Compensation (if shared)
+   - Skills & Certifications
+   - Education & Learning
+   - Personal Branding & Online Presence
+   - Achievements & Awards
+   - Social Proof & Networking
+   - Project Contributions & Leadership
+   - Work Performance & Impact Metrics
 
-EXTRACTION_PROMPT = """You are LOSS DOG's data processing system. Analyze the provided conversation and proactively interpret them for possible duduced 
-information that can be extracted. Turn the structured information into the following categories as json:
-
-1. Work History & Experience: Job titles, companies, industries, locations, adaptability, promotions
-2. Salary & Compensation: Base salary, bonuses, equity, benefits (if shared)
-3. Skills & Certifications: Technical skills, languages, certifications, licenses
-4. Education & Learning: Degrees, institutions, courses, research
-5. Personal Branding: Online presence, portfolio, blockchain projects, social media
-6. Achievements & Awards: Industry recognition, hackathons, creative projects
-7. Social Proof: Mentors, references, memberships, conferences
-8. Project Contributions: Major projects, open-source, patents, impact
-9. Performance Metrics: KPIs, revenue impact, growth metrics
-
-Format the output as clean, structured JSON.
-Mark any inferred information clearly."""
+3. Clean and structure the information:
+   - Deduplicate repeated information
+   - Resolve any inconsistencies
+   - Make reasonable inferences when dates or details are partial
+   - Standardize formatting (dates, company names, titles)
+
+4. Output a VALID JSON object with this exact structure:
+{
+    "work_history_experience": {
+        "positions": [
+            {
+                "title": "cleaned job title",
+                "company": "cleaned company name",
+                "duration": "standardized duration",
+                "description": "cleaned description",
+                "confidence": 0.95,
+                "inferred": false
+            }
+        ]
+    },
+    "skills_certifications": {
+        "technical_skills": ["skill1", "skill2"],
+        "certifications": [
+            {
+                "name": "certification name",
+                "issuer": "issuing organization",
+                "date": "YYYY-MM",
+                "confidence": 0.9
+            }
+        ]
+    }
+    // ... other categories following similar structure
+}
+
+IMPORTANT:
+- Return ONLY valid JSON
+- Always include confidence scores (0.0-1.0)
+- Mark any inferred information
+- Use consistent date formats (YYYY-MM-DD)
+- Clean and standardize all text fields
+- Return empty arrays [] for missing sections rather than null
+
+Example conversation snippet:
+User: "I worked at Google for a few years"
+Assistant: "That's interesting! What was your role there?"
+User: "I was a senior engineer, mostly doing ML stuff"
+
+Should extract to:
+{
+    "work_history_experience": {
+        "positions": [
+            {
+                "title": "Senior ML Engineer",
+                "company": "Google",
+                "duration": {
+                    "start": null,
+                    "end": null,
+                    "description": "multiple years",
+                    "inferred": true
+                },
+                "description": "Machine learning engineering",
+                "confidence": 0.85
+            }
+        ]
+    }
+}"""
 class ProfileBuilder:
     def __init__(self):
         self.conversation_history = []