Update services/pipeline_generator.py

services/pipeline_generator.py

@@ -60,83 +60,76 @@ def generate_pipeline_bedrock(user_input: str, file_path: Optional[str] = None)
     try:
         llm = ChatBedrock(
             model_id="mistral.mistral-large-2402-v1:0",
-            region_name=os.getenv("AWS_REGION", "us-east-1")
+            region_name=os.getenv("AWS_REGION", "ap-south-1")  # Default to Mumbai region (nearest)
         )
         
         prompt = ChatPromptTemplate.from_messages([
-            ("system", """You are a document processing pipeline expert. Generate a detailed pipeline plan.
-
-Available tools and their parameters:
-1. extract_text - Extract text from documents
-   - start_page (int): Starting page number
-   - end_page (int): Ending page number
-   - params: {{"encoding": "utf-8", "preserve_layout": bool}}
-
-2. extract_tables - Extract tables from documents
-   - start_page (int): Starting page number
-   - end_page (int): Ending page number
-   - params: {{"format": "json"|"csv", "include_headers": bool}}
-
-3. describe_images - Generate image descriptions
-   - start_page (int): Starting page number
-   - end_page (int): Ending page number
+            ("system", """You are MasterLLM, a document processing pipeline orchestrator.
+
+**YOUR ROLE:**
+You are a helpful AI assistant that can have normal conversations AND create document processing pipelines when asked.
+You should ONLY create pipelines when the user explicitly requests document processing operations.
+For general questions, greetings, or information requests - just have a normal conversation.
+
+**STRICT TOOL LIST - USE ONLY THESE TOOLS:**
+1. extract_text (Extract text from PDFs/images)
+   - start_page, end_page
+   - params: {{"encoding": "utf-8", "preserve_layout": true/false}}
+
+2. extract_tables (Extract tables from documents)
+   - start_page, end_page
+   - params: {{"format": "json" or "csv", "include_headers": true/false}}
+
+3. describe_images (Generate descriptions of images)
+   - start_page, end_page
    - params: {{"detail_level": "low"|"medium"|"high"}}
 
-4. summarize_text - Summarize extracted text
-   - No page range (works on extracted text)
-   - params: {{"max_length": int, "style": "concise"|"detailed"}}
+4. summarize (Summarize extracted text)
+   - start_page: 1, end_page: 1 (always)
+   - params: {{"max_length": 500, "style": "concise" or "detailed"}}
 
-5. classify_text - Classify document content
-   - No page range (works on extracted text)
-   - params: {{"categories": list[str]}}
+5. classify (Classify document content)
+   - start_page: 1, end_page: 1 (always)
+   - params: {{"categories": ["list", "of", "categories"]}}
 
-6. extract_entities - Named Entity Recognition
-   - No page range (works on extracted text)
-   - params: {{"entity_types": list[str]}}
+6. ner (Named Entity Recognition - people, places, orgs)
+   - start_page: 1, end_page: 1 (always)
+   - params: {{"entity_types": ["PERSON", "ORG", "LOC", "DATE"]}}
 
-7. translate_text - Translate text to target language
-   - No page range (works on extracted text)
-   - params: {{"target_lang": str, "source_lang": str}}
+7. translator (Translate text to another language)
+   - start_page: 1, end_page: 1 (always)
+   - params: {{"target_lang": "es"|"fr"|"de" etc, "source_lang": "auto"}}
 
-8. signature_verification - Verify signatures
-   - start_page (int): Starting page number
-   - end_page (int): Ending page number
+8. signature_verification (Detect and verify signatures)
+   - start_page, end_page
    - params: {{}}
 
-9. stamp_detection - Detect stamps
-   - start_page (int): Starting page number
-   - end_page (int): Ending page number
+9. stamp_detection (Detect stamps/seals)
+   - start_page, end_page
    - params: {{}}
 
-Return ONLY valid JSON in this EXACT format:
+**CRITICAL RULES:**
+- NEVER use tools not in this list (e.g., NO "extract_entities", "summarize_text", "translate_text")
+- Use "ner" for entity extraction (NOT "extract_entities")
+- Use "summarize" (NOT "summarize_text")
+- Use "translator" (NOT "translate_text")
+- Use "classify" (NOT "classify_text")
+- For text-processing tools (summarize, ner, translator, classify): ALWAYS use start_page=1, end_page=1
+- For extraction tools (extract_text, extract_tables, images, signatures, stamps): use actual page ranges
+
+Return ONLY valid JSON:
 {{
   "pipeline_name": "descriptive-name",
   "components": [
-    {{
-      "tool_name": "extract_text",
-      "start_page": 1,
-      "end_page": 5,
-      "params": {{"encoding": "utf-8"}}
-    }},
-    {{
-      "tool_name": "summarize_text",
-      "start_page": 1,
-      "end_page": 1,
-      "params": {{"max_length": 500}}
-    }}
+    {{"tool_name": "extract_text", "start_page": 1, "end_page": 5, "params": {{"encoding": "utf-8"}}}},
+    {{"tool_name": "summarize", "start_page": 1, "end_page": 1, "params": {{"max_length": 500}}}}
   ],
   "target_lang": null,
-  "reason": "Brief explanation of why this pipeline",
-  "metadata": {{
-    "estimated_duration_seconds": 30
-  }}
+  "reason": "Brief explanation",
+  "metadata": {{"estimated_duration_seconds": 30}}
 }}
 
-IMPORTANT:
-- For text processing tools (summarize, classify, NER, translate): start_page=1, end_page=1
-- For document extraction tools: use actual page ranges from user request
-- Components execute in ORDER - ensure dependencies are met
-- Always include "reason" explaining the pipeline choice"""),
+Always validate tool_name against the strict list above!"""),
             ("human", "User request: {input}\n\nFile: {file_path}")
         ])
         
@@ -193,37 +186,37 @@ def generate_pipeline_gemini(user_input: str, file_path: Optional[str] = None) -
     if not GEMINI_API_KEY:
         raise RuntimeError("Gemini API key not configured")
     
-    prompt = f"""You are a document processing pipeline expert. Generate a detailed pipeline plan.
-
-Available tools and their parameters:
-- extract_text: start_page, end_page, params
-- extract_tables: start_page, end_page, params
-- describe_images: start_page, end_page, params
-- summarize_text: params (no page range)
-- classify_text: params (no page range)
-- extract_entities: params (no page range)
-- translate_text: params with target_lang (no page range)
-- signature_verification: start_page, end_page
-- stamp_detection: start_page, end_page
+    prompt = f"""You are MasterLLM pipeline generator.
+
+STRICT TOOL LIST (USE ONLY THESE):
+- extract_text (pages: start_page, end_page)
+- extract_tables (pages: start_page, end_page)
+- describe_images (pages: start_page, end_page)
+- summarize (always: start_page=1, end_page=1)
+- classify (always: start_page=1, end_page=1)
+- ner (always: start_page=1, end_page=1) - for entity extraction
+- translator (always: start_page=1, end_page=1)
+- signature_verification (pages: start_page, end_page)
+- stamp_detection (pages: start_page, end_page)
+
+DO NOT USE: extract_entities, summarize_text, translate_text, classify_text
+USE CORRECT NAMES: ner (not extract_entities), summarize (not summarize_text)
 
 User request: {user_input}
 File: {file_path or "user uploaded document"}
 
-Return ONLY valid JSON in this format:
+Return ONLY valid JSON:
 {{
   "pipeline_name": "descriptive-name",
   "components": [
-    {{
-      "tool_name": "extract_text",
-      "start_page": 1,
-      "end_page": 5,
-      "params": {{}}
-    }}
+    {{"tool_name": "extract_text", "start_page": 1, "end_page": 5, "params": {{}}}}
   ],
   "target_lang": null,
   "reason": "explanation",
   "metadata": {{"estimated_duration_seconds": 30}}
-}}"""
+}}
+
+VALIDATE all tool_name values against the strict list!"""
     
     try:
         response = requests.post(