Sync from GitHub: 2086b6f3d02fad428eb1d39009f63d0130a06a64

inference.py

@@ -65,28 +65,37 @@ Output rules:
 
 # Two-step Chain of Thought prompts (reasoning mode) - OPTIMIZED FOR SPEED
 REASONING_PROMPT = """
-Analyze this Indian tractor invoice. Observe WITHOUT extracting:
+Analyze this Indian tractor invoice and share your observations about extracting these 4 fields:
+1. dealer_name
+2. model_name  
+3. horse_power
+4. asset_cost
 
-1. DEALER: Location, language, exact text
-2. MODEL: Location, format (checkbox/text), selected option, exact text
-3. HP: Location, format, selected value if checkbox, exact text
-4. TOTAL: Location, label, final amount with currency
-5. CHECKBOXES: Present? Which marked?
-6. CHALLENGES: Unclear handwriting or ambiguities?
+Think through each field:
 
-Return ONLY valid JSON:
-{
-  "dealer_text": string,
-  "model_text": string,
-  "model_is_checkbox": boolean,
-  "hp_text": string,
-  "hp_is_checkbox": boolean,
-  "amount_text": string,
-  "checkboxes_present": boolean,
-  "notes": string
-}
+DEALER NAME:
+- Where do you see the dealer/company name? Describe location and appearance,generally it is at top middle
+- What language is it in? Any spelling issues?
+- Write what you observe exactly as you see it
+
+MODEL NAME:
+- How is the model presented? (checkbox list, handwritten field, printed text)
+- If checkboxes exist, which one is marked/selected?
+- What exact text do you see for the model?
+- Is it in English or regional language?
+
+HORSE POWER:
+- Where do you see HP mentioned? 
+- Is it explicit (like "49 HP") or in a checkbox list?
+- If checkboxes, which HP value is selected?
+- What exact text shows the HP?
+
+ASSET COST:
+- Where is the final total amount located?
+- Which number is the final amount after all taxes?
+- What is the exact amount with currency symbol as shown?
 
-Preserve original language. Be concise.
+Express your observations naturally. Be specific about what you see and any uncertainties.
 """