Create document_generator_v2.py

document_generator_v2.py

@@ -0,0 +1,639 @@
+# File: prompts.py
+
+DOCUMENT_OUTLINE_PROMPT_SYSTEM = """You are a document generator. Provide the outline of the document requested in <prompt></prompt> in JSON format.
+Include sections and subsections if required. Use the "Content" field to provide a specific prompt or instruction for generating content for that particular section or subsection.
+make sure the Sections follow a logical flow and each prompt's content does not overlap with other sections.
+OUTPUT IN FOLLOWING JSON FORMAT enclosed in <output> tags
+<output>
+{
+    "Document": {
+      "Title": "Document Title",
+      "Author": "Author Name",
+      "Date": "YYYY-MM-DD",
+      "Version": "1.0",
+
+      "Sections": [
+        {
+          "SectionNumber": "1",
+          "Title": "Section Title",
+          "Content": "Specific prompt or instruction for generating content for this section",
+          "Subsections": [
+            {
+              "SectionNumber": "1.1",
+              "Title": "Subsection Title",
+              "Content": "Specific prompt or instruction for generating content for this subsection"
+            }
+          ]
+        }
+      ]
+    }
+  }
+</output>"""
+
+DOCUMENT_OUTLINE_PROMPT_USER = """<prompt>{query}</prompt>"""
+
+DOCUMENT_SECTION_PROMPT_SYSTEM = """You are a document generator, You need to output only the content requested in the section in the prompt.
+FORMAT YOUR OUTPUT AS MARKDOWN ENCLOSED IN <response></response> tags
+<overall_objective>{overall_objective}</overall_objective>
+<document_layout>{document_layout}</document_layout>"""
+
+DOCUMENT_SECTION_PROMPT_USER = """<prompt>Output the content for the section "{section_or_subsection_title}" formatted as markdown. Follow this instruction: {content_instruction}</prompt>"""
+
+##########################################
+
+DOCUMENT_TEMPLATE_OUTLINE_PROMPT_SYSTEM = """You are a document template generator. Provide the outline of the document requested in <prompt></prompt> in JSON format.
+Include sections and subsections if required. Use the "Content" field to provide a specific prompt or instruction for generating template with placeholder text /example content for that particular section or subsection. Specify in each prompt to output as a template and use placeholder text/ tables as necessory.
+make sure the Sections follow a logical flow and each prompt's content does not overlap with other sections.
+OUTPUT IN FOLLOWING JSON FORMAT enclosed in <output> tags
+<output>
+{
+    "Document": {
+      "Title": "Document Title",
+      "Author": "Author Name",
+      "Date": "YYYY-MM-DD",
+      "Version": "1.0",
+
+      "Sections": [
+        {
+          "SectionNumber": "1",
+          "Title": "Section Title",
+          "Content": "Specific prompt or instruction for generating template for this section",
+          "Subsections": [
+            {
+              "SectionNumber": "1.1",
+              "Title": "Subsection Title",
+              "Content": "Specific prompt or instruction for generating template for this subsection"
+            }
+          ]
+        }
+      ]
+    }
+  }
+</output>"""
+
+DOCUMENT_TEMPLATE_PROMPT_USER = """<prompt>{query}</prompt>"""
+
+DOCUMENT_TEMPLATE_SECTION_PROMPT_SYSTEM = """You are a document template generator,You need to output only the content requested in the section in the prompt, Use placeholder text/examples/tables wherever required.
+FORMAT YOUR OUTPUT AS A TEMPLATE ENCLOSED IN <response></response> tags
+<overall_objective>{overall_objective}</overall_objective>
+<document_layout>{document_layout}</document_layout>"""
+
+DOCUMENT_TEMPLATE_SECTION_PROMPT_USER = """<prompt>Output the content for the section "{section_or_subsection_title}" formatted as markdown. Follow this instruction: {content_instruction}</prompt>"""
+
+
+# File: llm_observability.py
+
+import sqlite3
+import json
+from datetime import datetime
+from typing import Dict, Any, List, Optional
+
+class LLMObservabilityManager:
+    def __init__(self, db_path: str = "llm_observability.db"):
+        self.db_path = db_path
+        self.create_table()
+
+    def create_table(self):
+        with sqlite3.connect(self.db_path) as conn:
+            cursor = conn.cursor()
+            cursor.execute('''
+                CREATE TABLE IF NOT EXISTS llm_observations (
+                    id TEXT PRIMARY KEY,
+                    conversation_id TEXT,
+                    created_at DATETIME,
+                    status TEXT,
+                    request TEXT,
+                    response TEXT,
+                    model TEXT,
+                    total_tokens INTEGER,
+                    prompt_tokens INTEGER,
+                    completion_tokens INTEGER,
+                    latency FLOAT,
+                    user TEXT
+                )
+            ''')
+
+    def insert_observation(self, response: Dict[str, Any], conversation_id: str, status: str, request: str, latency: float, user: str):
+        created_at = datetime.fromtimestamp(response['created'])
+        
+        with sqlite3.connect(self.db_path) as conn:
+            cursor = conn.cursor()
+            cursor.execute('''
+                INSERT INTO llm_observations 
+                (id, conversation_id, created_at, status, request, response, model, total_tokens, prompt_tokens, completion_tokens, latency, user)
+                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+            ''', (
+                response['id'],
+                conversation_id,
+                created_at,
+                status,
+                request,
+                json.dumps(response['choices'][0]['message']),
+                response['model'],
+                response['usage']['total_tokens'],
+                response['usage']['prompt_tokens'],
+                response['usage']['completion_tokens'],
+                latency,
+                user
+            ))
+
+    def get_observations(self, conversation_id: Optional[str] = None) -> List[Dict[str, Any]]:
+        with sqlite3.connect(self.db_path) as conn:
+            cursor = conn.cursor()
+            if conversation_id:
+                cursor.execute('SELECT * FROM llm_observations WHERE conversation_id = ? ORDER BY created_at', (conversation_id,))
+            else:
+                cursor.execute('SELECT * FROM llm_observations ORDER BY created_at')
+            rows = cursor.fetchall()
+
+            column_names = [description[0] for description in cursor.description]
+            return [dict(zip(column_names, row)) for row in rows]
+
+    def get_all_observations(self) -> List[Dict[str, Any]]:
+        return self.get_observations()
+
+
+# File: app.py
+import os
+import json
+import re
+import asyncio
+import time
+from typing import List, Dict, Optional, Any, Callable, Union
+from openai import OpenAI
+import logging
+import functools
+from fastapi import APIRouter, HTTPException, Request, UploadFile, File, Depends
+from fastapi.responses import StreamingResponse
+from pydantic import BaseModel
+from fastapi_cache.decorator import cache
+import psycopg2
+from datetime import datetime
+import base64
+from fastapi import Form 
+
+logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
+logger = logging.getLogger(__name__)
+
+def log_execution(func: Callable) -> Callable:
+    @functools.wraps(func)
+    def wrapper(*args: Any, **kwargs: Any) -> Any:
+        logger.info(f"Executing {func.__name__}")
+        try:
+            result = func(*args, **kwargs)
+            logger.info(f"{func.__name__} completed successfully")
+            return result
+        except Exception as e:
+            logger.error(f"Error in {func.__name__}: {e}")
+            raise
+    return wrapper
+
+# aiclient.py
+
+class AIClient:
+    def __init__(self):
+        self.client = OpenAI(
+            base_url="https://openrouter.ai/api/v1",
+            api_key="sk-or-v1-" + os.environ['OPENROUTER_API_KEY']
+        )
+        self.observability_manager = LLMObservabilityManager()
+
+    @log_execution
+    def generate_response(
+        self,
+        messages: List[Dict[str, str]],
+        model: str = "openai/gpt-4o-mini",
+        max_tokens: int = 32000,
+        conversation_id: str = None,
+        user: str = "anonymous"
+    ) -> Optional[str]:
+        if not messages:
+            return None
+        
+        start_time = time.time()
+        response = self.client.chat.completions.create(
+            model=model,
+            messages=messages,
+            max_tokens=max_tokens,
+            stream=False
+        )
+        end_time = time.time()
+        latency = end_time - start_time
+
+        # Log the observation
+        self.observability_manager.insert_observation(
+            response=response.dict(),
+            conversation_id=conversation_id or "default",
+            status="success",
+            request=json.dumps(messages),
+            latency=latency,
+            user=user
+        )
+
+        return response.choices[0].message.content
+
+    @log_execution
+    def generate_vision_response(
+        self,
+        messages: List[Dict[str, Union[str, List[Dict[str, Union[str, Dict[str, str]]]]]]],
+        model: str = "google/gemini-flash-1.5-8b",
+        max_tokens: int = 32000,
+        conversation_id: str = None,
+        user: str = "anonymous"
+    ) -> Optional[str]:
+        if not messages:
+            return None
+        
+        start_time = time.time()
+        response = self.client.chat.completions.create(
+            model=model,
+            messages=messages,
+            max_tokens=max_tokens,
+            stream=False
+        )
+        end_time = time.time()
+        latency = end_time - start_time
+
+        # Log the observation
+        self.observability_manager.insert_observation(
+            response=response.dict(),
+            conversation_id=conversation_id or "default",
+            status="success",
+            request=json.dumps(messages),
+            latency=latency,
+            user=user
+        )
+
+        return response.choices[0].message.content
+
+
+class VisionTools:
+    def __init__(self, ai_client):
+        self.ai_client = ai_client
+
+    async def extract_images_info(self, images: List[UploadFile]) -> str:
+        try:
+            image_contents = []
+            for image in images:
+                image_content = await image.read()
+                base64_image = base64.b64encode(image_content).decode('utf-8')
+                image_contents.append({
+                    "type": "image_url",
+                    "image_url": {
+                        "url": f"data:image/jpeg;base64,{base64_image}"
+                    }
+                })
+            
+            messages = [
+                {
+                    "role": "user",
+                    "content": [
+                        {
+                            "type": "text",
+                            "text": "Extract the contents of these images in detail in a structured format, focusing on any text, tables, diagrams, or visual elements that might be relevant for document generation."
+                        },
+                        *image_contents
+                    ]
+                }
+            ]
+            
+            image_context = self.ai_client.generate_vision_response(messages)
+            return image_context
+        except Exception as e:
+            print(f"Error processing images: {str(e)}")
+            return ""
+
+
+class DatabaseManager:
+    """Manages database operations."""
+
+    def __init__(self):
+        self.db_params = {
+            "dbname": "postgres",
+            "user": os.environ['SUPABASE_USER'],
+            "password": os.environ['SUPABASE_PASSWORD'],
+            "host": "aws-0-us-west-1.pooler.supabase.com",
+            "port": "5432"
+        }
+
+    @log_execution
+    def update_database(self, user_id: str, user_query: str, response: str) -> None:
+        with psycopg2.connect(**self.db_params) as conn:
+            with conn.cursor() as cur:
+                insert_query = """
+                INSERT INTO ai_document_generator (user_id, user_query, response)
+                VALUES (%s, %s, %s);
+                """
+                cur.execute(insert_query, (user_id, user_query, response))
+        
+class DocumentGenerator:
+    def __init__(self, ai_client: AIClient):
+        self.ai_client = ai_client
+        self.document_outline = None
+        self.content_messages = []
+
+    @staticmethod
+    def extract_between_tags(text: str, tag: str) -> str:
+        pattern = f"<{tag}>(.*?)</{tag}>"
+        match = re.search(pattern, text, re.DOTALL)
+        return match.group(1).strip() if match else ""
+
+    @staticmethod
+    def remove_duplicate_title(content: str, title: str, section_number: str) -> str:
+        patterns = [
+            rf"^#+\s*{re.escape(section_number)}(?:\s+|\s*:\s*|\.\s*){re.escape(title)}",
+            rf"^#+\s*{re.escape(title)}",
+            rf"^{re.escape(section_number)}(?:\s+|\s*:\s*|\.\s*){re.escape(title)}",
+            rf"^{re.escape(title)}",
+        ]
+        
+        for pattern in patterns:
+            content = re.sub(pattern, "", content, flags=re.MULTILINE | re.IGNORECASE)
+        
+        return content.lstrip()
+
+    @log_execution
+    def generate_document_outline(self, query: str, template: bool = False, image_context: str = "", max_retries: int = 3) -> Optional[Dict]:
+        messages = [
+            {"role": "system", "content": DOCUMENT_OUTLINE_PROMPT_SYSTEM if not template else DOCUMENT_TEMPLATE_OUTLINE_PROMPT_SYSTEM},
+            {"role": "user", "content": DOCUMENT_OUTLINE_PROMPT_USER.format(query=query) if not template else DOCUMENT_TEMPLATE_PROMPT_USER.format(query=query, image_context=image_context)}
+        ]
+        # Update user content to include image context if provided
+        if image_context:
+            messages[1]["content"] += f"<attached_images>\n\n{image_context}\n\n</attached_images>"
+        
+        for attempt in range(max_retries):
+            outline_response = self.ai_client.generate_response(messages, model="openai/gpt-4o")
+            outline_json_text = self.extract_between_tags(outline_response, "output")
+            
+            try:
+                self.document_outline = json.loads(outline_json_text)
+                return self.document_outline
+            except json.JSONDecodeError as e:
+                if attempt < max_retries - 1:
+                    logger.warning(f"Failed to parse JSON (attempt {attempt + 1}): {e}")
+                    logger.info("Retrying...")
+                else:
+                    logger.error(f"Failed to parse JSON after {max_retries} attempts: {e}")
+                    return None
+
+    @log_execution
+    def generate_content(self, title: str, content_instruction: str, section_number: str, template: bool = False) -> str:
+        SECTION_PROMPT_USER = DOCUMENT_SECTION_PROMPT_USER if not template else DOCUMENT_TEMPLATE_SECTION_PROMPT_USER
+        self.content_messages.append({
+            "role": "user",
+            "content": SECTION_PROMPT_USER.format(
+                section_or_subsection_title=title,
+                content_instruction=content_instruction
+            )
+        })
+        section_response = self.ai_client.generate_response(self.content_messages)
+        content = self.extract_between_tags(section_response, "response")
+        content = self.remove_duplicate_title(content, title, section_number)
+        self.content_messages.append({
+            "role": "assistant",
+            "content": section_response
+        })
+        return content
+
+class MarkdownConverter:
+    @staticmethod
+    def slugify(text: str) -> str:
+        return re.sub(r'\W+', '-', text.lower())
+
+    @classmethod
+    def generate_toc(cls, sections: List[Dict]) -> str:
+        toc = "<div style='page-break-before: always;'></div>\n\n"
+        toc += "<h2 style='color: #2c3e50; text-align: center;'>Table of Contents</h2>\n\n"
+        toc += "<nav style='background-color: #f8f9fa; padding: 20px; border-radius: 5px; line-height: 1.6;'>\n\n"
+        for section in sections:
+            section_number = section['SectionNumber']
+            section_title = section['Title']
+            toc += f"<p><a href='#{cls.slugify(section_title)}' style='color: #3498db; text-decoration: none;'>{section_number}. {section_title}</a></p>\n\n"
+            
+            for subsection in section.get('Subsections', []):
+                subsection_number = subsection['SectionNumber']
+                subsection_title = subsection['Title']
+                toc += f"<p style='margin-left: 20px;'><a href='#{cls.slugify(subsection_title)}' style='color: #2980b9; text-decoration: none;'>{subsection_number} {subsection_title}</a></p>\n\n"
+        
+        toc += "</nav>\n\n"
+        return toc
+
+    @classmethod
+    def convert_to_markdown(cls, document: Dict) -> str:
+        markdown = "<div style='text-align: center; padding-top: 33vh;'>\n\n"
+        markdown += f"<h1 style='color: #2c3e50; border-bottom: 2px solid #3498db; padding-bottom: 10px; display: inline-block;'>{document['Title']}</h1>\n\n"
+        markdown += f"<p style='color: #7f8c8d;'><em>By {document['Author']}</em></p>\n\n"
+        markdown += f"<p style='color: #95a5a6;'>Version {document['Version']} | {document['Date']}</p>\n\n"
+        markdown += "</div>\n\n"
+        
+        markdown += cls.generate_toc(document['Sections'])
+        
+        markdown += "<div style='max-width: 800px; margin: 0 auto; font-family: \"Segoe UI\", Arial, sans-serif; line-height: 1.6;'>\n\n"
+        
+        for section in document['Sections']:
+            markdown += "<div style='page-break-before: always;'></div>\n\n"
+            section_number = section['SectionNumber']
+            section_title = section['Title']
+            markdown += f"<h2 id='{cls.slugify(section_title)}' style='color: #2c3e50; border-bottom: 1px solid #bdc3c7; padding-bottom: 5px;'>{section_number}. {section_title}</h2>\n\n"
+            markdown += f"<div style='color: #34495e; margin-bottom: 20px;'>\n\n{section['Content']}\n\n</div>\n\n"
+            
+            for subsection in section.get('Subsections', []):
+                subsection_number = subsection['SectionNumber']
+                subsection_title = subsection['Title']
+                markdown += f"<h3 id='{cls.slugify(subsection_title)}' style='color: #34495e;'>{subsection_number} {subsection_title}</h3>\n\n"
+                markdown += f"<div style='color: #34495e; margin-bottom: 20px;'>\n\n{subsection['Content']}\n\n</div>\n\n"
+        
+        markdown += "</div>"
+        return markdown
+
+router = APIRouter()
+
+class JsonDocumentResponse(BaseModel):
+    json_document: Dict
+
+# class JsonDocumentRequest(BaseModel):
+#     query: str
+#     template: bool = False
+#     images: Optional[List[UploadFile]] = File(None)
+#     documents: Optional[List[UploadFile]] = File(None)
+#     conversation_id: str = ""
+
+class MarkdownDocumentRequest(BaseModel):
+    json_document: Dict
+    query: str
+
+MESSAGE_DELIMITER = b"\n---DELIMITER---\n"
+
+def yield_message(message):
+    message_json = json.dumps(message, ensure_ascii=False).encode('utf-8')
+    return message_json + MESSAGE_DELIMITER
+
+async def generate_document_stream(document_generator: DocumentGenerator, document_outline: Dict, query: str, template: bool = False):
+    document_generator.document_outline = document_outline
+    db_manager = DatabaseManager()
+    overall_objective = query
+    document_layout = json.dumps(document_generator.document_outline, indent=2)
+
+    SECTION_PROMPT_SYSTEM = DOCUMENT_SECTION_PROMPT_SYSTEM if not template else DOCUMENT_TEMPLATE_SECTION_PROMPT_SYSTEM
+    document_generator.content_messages = [
+        {
+            "role": "system",
+            "content": SECTION_PROMPT_SYSTEM.format(
+                overall_objective=overall_objective,
+                document_layout=document_layout
+            )
+        }
+    ]
+
+    for section in document_generator.document_outline["Document"].get("Sections", []):
+        section_title = section.get("Title", "")
+        section_number = section.get("SectionNumber", "")
+        content_instruction = section.get("Content", "")
+        logging.info(f"Generating content for section: {section_title}")
+        content = document_generator.generate_content(section_title, content_instruction, section_number, template)
+        section["Content"] = content
+        yield yield_message({
+            "type": "document_section",
+            "content": {
+                "section_number": section_number,
+                "section_title": section_title,
+                "content": content
+            }
+        })
+
+        for subsection in section.get("Subsections", []):
+            subsection_title = subsection.get("Title", "")
+            subsection_number = subsection.get("SectionNumber", "")
+            subsection_content_instruction = subsection.get("Content", "")
+            logging.info(f"Generating content for subsection: {subsection_title}")
+            content = document_generator.generate_content(subsection_title, subsection_content_instruction, subsection_number, template)
+            subsection["Content"] = content
+            yield yield_message({
+                "type": "document_section",
+                "content": {
+                    "section_number": subsection_number,
+                    "section_title": subsection_title,
+                    "content": content
+                }
+            })
+
+    markdown_document = MarkdownConverter.convert_to_markdown(document_generator.document_outline["Document"])
+    
+    yield yield_message({
+        "type": "complete_document",
+        "content": {
+            "markdown": markdown_document,
+            "json": document_generator.document_outline
+                    },
+            });
+
+    db_manager.update_database("elevatics", query, markdown_document)
+
+@router.post("/generate-document/markdown-stream")
+async def generate_markdown_document_stream_endpoint(request: MarkdownDocumentRequest):
+    ai_client = AIClient()
+    document_generator = DocumentGenerator(ai_client)
+    
+    async def stream_generator():
+        try:
+            async for chunk in generate_document_stream(document_generator, request.json_document, request.query, request.template):
+                yield chunk
+        except Exception as e:
+            yield yield_message({
+                "type": "error",
+                "content": str(e)
+            })
+
+    return StreamingResponse(stream_generator(), media_type="application/octet-stream")
+
+
+@cache(expire=600*24*7)
+@router.post("/generate-document/json", response_model=JsonDocumentResponse)
+async def generate_document_outline_endpoint(
+    query: str = Form(...),
+    template: bool = Form(False),
+    conversation_id: str = Form(...),
+    images: Optional[List[UploadFile]] = File(None),
+    documents: Optional[List[UploadFile]] = File(None)
+):
+    ai_client = AIClient()
+    document_generator = DocumentGenerator(ai_client)
+    vision_tools = VisionTools(ai_client)
+    
+    try:
+        image_context = ""
+        if images:
+            image_context = await vision_tools.extract_images_info(images)
+        
+        json_document = document_generator.generate_document_outline(
+            query,
+            template,
+            image_context=image_context
+        )
+        
+        if json_document is None:
+            raise HTTPException(status_code=500, detail="Failed to generate a valid document outline")
+        
+        return JsonDocumentResponse(json_document=json_document)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+
+## OBSERVABILITY
+from uuid import uuid4
+import csv
+from io import StringIO
+
+class ObservationResponse(BaseModel):
+    observations: List[Dict]
+    
+def create_csv_response(observations: List[Dict]) -> StreamingResponse:
+    def iter_csv(data):
+        output = StringIO()
+        writer = csv.DictWriter(output, fieldnames=data[0].keys() if data else [])
+        writer.writeheader()
+        for row in data:
+            writer.writerow(row)
+        output.seek(0)
+        yield output.read()
+
+    headers = {
+        'Content-Disposition': 'attachment; filename="observations.csv"'
+    }
+    return StreamingResponse(iter_csv(observations), media_type="text/csv", headers=headers)
+    
+
+@router.get("/last-observations/{limit}")
+async def get_last_observations(limit: int = 10, format: str = "json"):
+    observability_manager = LLMObservabilityManager()
+    
+    try:
+        # Get all observations, sorted by created_at in descending order
+        all_observations = observability_manager.get_observations()
+        all_observations.sort(key=lambda x: x['created_at'], reverse=True)
+        
+        # Get the last conversation_id
+        if all_observations:
+            last_conversation_id = all_observations[0]['conversation_id']
+            
+            # Filter observations for the last conversation
+            last_conversation_observations = [
+                obs for obs in all_observations
+                if obs['conversation_id'] == last_conversation_id
+            ][:limit]
+            
+            if format.lower() == "csv":
+                return create_csv_response(last_conversation_observations)
+            else:
+                return ObservationResponse(observations=last_conversation_observations)
+        else:
+            if format.lower() == "csv":
+                return create_csv_response([])
+            else:
+                return ObservationResponse(observations=[])
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Failed to retrieve observations: {str(e)}")
+
+