# File: prompts.py DOCUMENT_OUTLINE_PROMPT_SYSTEM = """You are a document generator. Provide the outline of the document requested in 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 tags { "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" } ] } ] } } """ DOCUMENT_OUTLINE_PROMPT_USER = """{query}""" 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 tags {overall_objective} {document_layout}""" DOCUMENT_SECTION_PROMPT_USER = """Output the content for the section "{section_or_subsection_title}" formatted as markdown. Follow this instruction: {content_instruction}""" ########################################## DOCUMENT_TEMPLATE_OUTLINE_PROMPT_SYSTEM = """You are a document template generator. Provide the outline of the document requested in 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 tags { "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" } ] } ] } } """ DOCUMENT_TEMPLATE_PROMPT_USER = """{query}""" 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 tags {overall_objective} {document_layout}""" DOCUMENT_TEMPLATE_SECTION_PROMPT_USER = """Output the content for the section "{section_or_subsection_title}" formatted as markdown. Follow this instruction: {content_instruction}""" # 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 from fastapi.responses import StreamingResponse from pydantic import BaseModel from fastapi_cache.decorator import cache import psycopg2 from datetime import datetime import base64 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}>(.*?)" 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, 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)} ] 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 = "
\n\n" toc += "

Table of Contents

\n\n" toc += "\n\n" return toc @classmethod def convert_to_markdown(cls, document: Dict) -> str: markdown = "
\n\n" markdown += f"

{document['Title']}

\n\n" markdown += f"

By {document['Author']}

\n\n" markdown += f"

Version {document['Version']} | {document['Date']}

\n\n" markdown += "
\n\n" markdown += cls.generate_toc(document['Sections']) markdown += "
\n\n" for section in document['Sections']: markdown += "
\n\n" section_number = section['SectionNumber'] section_title = section['Title'] markdown += f"

{section_number}. {section_title}

\n\n" markdown += f"
\n\n{section['Content']}\n\n
\n\n" for subsection in section.get('Subsections', []): subsection_number = subsection['SectionNumber'] subsection_title = subsection['Title'] markdown += f"

{subsection_number} {subsection_title}

\n\n" markdown += f"
\n\n{subsection['Content']}\n\n
\n\n" markdown += "
" return markdown router = APIRouter() class JsonDocumentResponse(BaseModel): json_document: Dict class MarkdownDocumentRequest(BaseModel): json_document: Dict query: str template: bool = False images: Optional[List[UploadFile]] = File(None) 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(request: MarkdownDocumentRequest): ai_client = AIClient() document_generator = DocumentGenerator(ai_client) vision_tools = VisionTools(ai_client) try: image_context = "" if request.images: image_context = await vision_tools.extract_images_info(request.images) json_document = document_generator.generate_document_outline( request.query, request.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)}") ########################################### class MarkdownDocumentResponse(BaseModel): markdown_document: str @router.post("/generate-document-test", response_model=MarkdownDocumentResponse) async def test_generate_document_endpoint(request: MarkdownDocumentRequest): try: # Load JSON document from file json_path = os.path.join("output/document_generator", "ai-chatbot-prd.json") with open(json_path, "r") as json_file: json_document = json.load(json_file) # Load Markdown document from file md_path = os.path.join("output/document_generator", "ai-chatbot-prd.md") with open(md_path, "r") as md_file: markdown_document = md_file.read() return MarkdownDocumentResponse(markdown_document=markdown_document) except FileNotFoundError: raise HTTPException(status_code=404, detail="Test files not found") except json.JSONDecodeError: raise HTTPException(status_code=500, detail="Error parsing JSON file") except Exception as e: raise HTTPException(status_code=500, detail=str(e)) class CacheTestResponse(BaseModel): result: str execution_time: float @router.get("/test-cache/{test_id}", response_model=CacheTestResponse) @cache(expire=60) # Cache for 1 minute async def test_cache(test_id: int): start_time = time.time() # Simulate some time-consuming operation await asyncio.sleep(2) result = f"Test result for ID: {test_id}" end_time = time.time() execution_time = end_time - start_time return CacheTestResponse( result=result, execution_time=execution_time )