equilink-site/server/prompt_chaining/step_3_create_posts.py
2025-02-12 17:38:06 +05:30

162 lines
5.7 KiB
Python

#
# Module: step_3
#
# This module implements the step_3 function for creating a new episode posts for an agent.
#
# Title: Step 3
# Summary: Step 3 implementation.
# Authors:
# - @TheBlockRhino
# Created: 2025-01-02
# Last edited by: @TheBlockRhino
# Last edited date: 2025-01-09
# URLs:
# - https://arai-ai.io
# - https://github.com/ARAI-DevHub/arai-ai-agents
# - https://x.com/TheBlockRhino
# standard imports
import os
import yaml
import json
import sys
import time
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
# custom ARAI imports
from utils.content_generator import ContentGenerator
from utils.template_types import TemplateType
def create_episode_posts(ai_model, master_file_path, number_of_posts):
'''
Description:
Create a new episode posts for the agent
Args:
ai_model: The AI model to use for generating responses
master_file_path: The path to the agent master json file
Raises:
Exception: If there's an error creating the episode posts
Example:
>>> ai_model = GeminiModel()
>>> master_file_path = "configs/Zorp/Zorp_master.json"
>>> step_3(ai_model, master_file_path)
'''
print("Step 3: Create a batch of posts for episode")
# step 3.1: load the season template yaml file
manager = ContentGenerator()
episode_template = manager.create_new_template_json(TemplateType.EPISODE)
# step 2.2: load the agent json file
agent_master_json = None
with open(master_file_path, 'r', encoding='utf-8') as file:
agent_master_json = json.load(file)
# extract agent details from master json
agent_details = agent_master_json['agent']['agent_details']
# get last season file
current_season = None
# Get the last season from the seasons array if it exists
seasons = agent_master_json['agent']['seasons']
current_season = seasons[-1] if seasons else None
# Extract season details
season_details = {
'season_name': current_season['season_name'],
'season_number': current_season['season_number'],
'season_description': current_season['season_description'],
'season_highlights': current_season['season_highlights'],
'season_summary': current_season['season_summary'],
}
previous_episode = None
# Iterate over episodes while maintaining season context
for episode in current_season['episodes']:
# Now you have access to both season_details and the current episode
print(f"Processing Episode {episode['episode_number']} of Season {season_details['season_number']}")
# Work with episode data
episode_data = {
'episode_name': episode['episode_name'],
'episode_number': episode['episode_number'],
'episode_description': episode['episode_description'],
'episode_highlights': episode['episode_highlights'],
'episode_summary': episode['episode_summary'],
}
# step 3.10: Generate a new season name, topic, and communication style with the prompt_2 template
# prompt 2 Season Creation:
# note that emojis will be output as unicode characters due to the yaml dump
prompt_3_vars = {
"agent_name": agent_details["name"],
"agent_json": json.dumps(agent_details),
"season_json": json.dumps(season_details),
"episode_json": json.dumps(episode_data),
"previous_episode": json.dumps(previous_episode),
"number_of_posts": number_of_posts,
"post_length": 277
}
# Constants for retry configuration
max_retries = 3
delay = 2 # seconds between retries
# step 3.11: Run the prompt with retry logic for LLM failures
success = False
for attempt in range(max_retries):
try:
posts_data = manager.run_prompt(
prompt_key="prompt_3 (Episode Posts Creation)",
template_vars=prompt_3_vars,
ai_model=ai_model,
)
# Validate that posts_data is valid JSON and has expected structure
if isinstance(posts_data, dict) and 'posts' in posts_data:
success = True
break
else:
print(f"Invalid response format. Attempt {attempt + 1}/{max_retries}")
except (json.JSONDecodeError, Exception) as e:
print(f"Error on attempt {attempt + 1}/{max_retries}: {str(e)}")
if attempt < max_retries - 1:
print(f"Retrying in {delay} seconds...")
time.sleep(delay)
continue
if not success:
print("Max retries reached. Failed to get valid response from LLM.")
return None
print("Appending the episode data to the master data")
agent_master_json = manager.append_episodes(
master_data=agent_master_json,
posts_data=posts_data,
season_index=season_details["season_number"]-1,
episode_index= episode_data["episode_number"]-1
)
# step 2.9: save the master data to a file
print("Saving the master data to a file")
manager.save_json_file(
save_path=master_file_path,
json_data=agent_master_json
)
# step 3.15: Move onto the next step of creating a new season
previous_episode = episode_data
import models.gemini_model as gemini_model
if __name__ == "__main__":
ai_model = gemini_model.GeminiModel()
create_episode_posts(ai_model, "configs/Pixelina/Pixelina_master.json", 6)