Files
daily-journal-prompt/backend/app/services/ai_service.py

353 lines
13 KiB
Python

"""
AI service for handling OpenAI/DeepSeek API calls.
"""
import json
from typing import List, Dict, Any, Optional
from openai import OpenAI, AsyncOpenAI
from app.core.config import settings
from app.core.logging import setup_logging
logger = setup_logging()
class AIService:
"""Service for handling AI API calls."""
def __init__(self):
"""Initialize AI service."""
api_key = settings.DEEPSEEK_API_KEY or settings.OPENAI_API_KEY
if not api_key:
raise ValueError("No API key found. Set DEEPSEEK_API_KEY or OPENAI_API_KEY in environment.")
self.client = AsyncOpenAI(
api_key=api_key,
base_url=settings.API_BASE_URL
)
self.model = settings.MODEL
def _clean_ai_response(self, response_content: str) -> str:
"""
Clean up AI response content to handle common formatting issues.
Handles:
1. Leading/trailing backticks (```json ... ```)
2. Leading "json" string on its own line
3. Extra whitespace and newlines
"""
content = response_content.strip()
# Remove leading/trailing backticks (```json ... ```)
if content.startswith('```'):
lines = content.split('\n')
if len(lines) > 1:
first_line = lines[0].strip()
if 'json' in first_line.lower() or first_line == '```':
content = '\n'.join(lines[1:])
# Remove trailing backticks if present
if content.endswith('```'):
content = content[:-3].rstrip()
# Remove leading "json" string on its own line (case-insensitive)
lines = content.split('\n')
if len(lines) > 0:
first_line = lines[0].strip().lower()
if first_line == 'json':
content = '\n'.join(lines[1:])
# Also handle the case where "json" might be at the beginning of the first line
content = content.strip()
if content.lower().startswith('json\n'):
content = content[4:].strip()
return content.strip()
async def generate_prompts(
self,
prompt_template: str,
historic_prompts: List[Dict[str, str]],
feedback_words: Optional[List[Dict[str, Any]]] = None,
count: Optional[int] = None,
min_length: Optional[int] = None,
max_length: Optional[int] = None
) -> List[str]:
"""
Generate journal prompts using AI.
Args:
prompt_template: Base prompt template
historic_prompts: List of historic prompts for context
feedback_words: List of feedback words with weights
count: Number of prompts to generate
min_length: Minimum prompt length
max_length: Maximum prompt length
Returns:
List of generated prompts
"""
if count is None:
count = settings.NUM_PROMPTS_PER_SESSION
if min_length is None:
min_length = settings.MIN_PROMPT_LENGTH
if max_length is None:
max_length = settings.MAX_PROMPT_LENGTH
# Prepare the full prompt
full_prompt = self._prepare_prompt(
prompt_template,
historic_prompts,
feedback_words,
count,
min_length,
max_length
)
logger.info(f"Generating {count} prompts with AI")
try:
# Call the AI API
response = await self.client.chat.completions.create(
model=self.model,
messages=[
{
"role": "system",
"content": "You are a creative writing assistant that generates journal prompts. Always respond with valid JSON."
},
{
"role": "user",
"content": full_prompt
}
],
temperature=0.7,
max_tokens=2000
)
response_content = response.choices[0].message.content
logger.debug(f"AI response received: {len(response_content)} characters")
# Parse the response
prompts = self._parse_prompt_response(response_content, count)
logger.info(f"Successfully parsed {len(prompts)} prompts from AI response")
return prompts
except Exception as e:
logger.error(f"Error calling AI API: {e}")
logger.debug(f"Full prompt sent to API: {full_prompt[:500]}...")
raise
def _prepare_prompt(
self,
template: str,
historic_prompts: List[Dict[str, str]],
feedback_words: Optional[List[Dict[str, Any]]],
count: int,
min_length: int,
max_length: int
) -> str:
"""Prepare the full prompt with all context."""
# Add the instruction for the specific number of prompts
prompt_instruction = f"Please generate {count} writing prompts, each between {min_length} and {max_length} characters."
# Start with template and instruction
full_prompt = f"{template}\n\n{prompt_instruction}"
# Add historic prompts if available
if historic_prompts:
historic_context = json.dumps(historic_prompts, indent=2)
full_prompt = f"{full_prompt}\n\nPrevious prompts:\n{historic_context}"
# Add feedback words if available
if feedback_words:
feedback_context = json.dumps(feedback_words, indent=2)
full_prompt = f"{full_prompt}\n\nFeedback words:\n{feedback_context}"
return full_prompt
def _parse_prompt_response(self, response_content: str, expected_count: int) -> List[str]:
"""Parse AI response to extract prompts."""
cleaned_content = self._clean_ai_response(response_content)
try:
data = json.loads(cleaned_content)
if isinstance(data, list):
if len(data) >= expected_count:
return data[:expected_count]
else:
logger.warning(f"AI returned {len(data)} prompts, expected {expected_count}")
return data
elif isinstance(data, dict):
logger.warning("AI returned dictionary format, expected list format")
prompts = []
for i in range(expected_count):
key = f"newprompt{i}"
if key in data:
prompts.append(data[key])
return prompts
else:
logger.warning(f"AI returned unexpected data type: {type(data)}")
return []
except json.JSONDecodeError:
logger.warning("AI response is not valid JSON, attempting to extract prompts...")
return self._extract_prompts_from_text(response_content, expected_count)
def _extract_prompts_from_text(self, text: str, expected_count: int) -> List[str]:
"""Extract prompts from plain text response."""
lines = text.strip().split('\n')
prompts = []
for line in lines[:expected_count]:
line = line.strip()
if line and len(line) > 50: # Reasonable minimum length for a prompt
prompts.append(line)
return prompts
async def generate_theme_feedback_words(
self,
feedback_template: str,
historic_prompts: List[Dict[str, str]],
queued_feedback_words: Optional[List[Dict[str, Any]]] = None,
historic_feedback_words: Optional[List[Dict[str, Any]]] = None
) -> List[str]:
"""
Generate theme feedback words using AI.
Args:
feedback_template: Feedback analysis template
historic_prompts: List of historic prompts for context
queued_feedback_words: Queued feedback words with weights (positions 0-5)
historic_feedback_words: Historic feedback words with weights (all positions)
Returns:
List of 6 theme words
"""
# Prepare the full prompt
full_prompt = self._prepare_feedback_prompt(
feedback_template,
historic_prompts,
queued_feedback_words,
historic_feedback_words
)
logger.info("Generating theme feedback words with AI")
try:
# Call the AI API
response = await self.client.chat.completions.create(
model=self.model,
messages=[
{
"role": "system",
"content": "You are a creative writing assistant that analyzes writing prompts. Always respond with valid JSON."
},
{
"role": "user",
"content": full_prompt
}
],
temperature=0.7,
max_tokens=1000
)
response_content = response.choices[0].message.content
logger.debug(f"AI feedback response received: {len(response_content)} characters")
# Parse the response
theme_words = self._parse_feedback_response(response_content)
logger.info(f"Successfully parsed {len(theme_words)} theme words from AI response")
if len(theme_words) != 6:
logger.warning(f"Expected 6 theme words, got {len(theme_words)}")
return theme_words
except Exception as e:
logger.error(f"Error calling AI API for feedback: {e}")
logger.debug(f"Full feedback prompt sent to API: {full_prompt[:500]}...")
raise
def _prepare_feedback_prompt(
self,
template: str,
historic_prompts: List[Dict[str, str]],
queued_feedback_words: Optional[List[Dict[str, Any]]],
historic_feedback_words: Optional[List[Dict[str, Any]]]
) -> str:
"""Prepare the full feedback prompt."""
if not historic_prompts:
raise ValueError("No historic prompts available for feedback analysis")
full_prompt = f"{template}\n\nPrevious prompts:\n{json.dumps(historic_prompts, indent=2)}"
# Add queued feedback words if available (these have user-adjusted weights)
if queued_feedback_words:
# Extract just the words and weights for clarity
queued_words_with_weights = []
for item in queued_feedback_words:
key = list(item.keys())[0]
word = item[key]
weight = item.get("weight", 3)
queued_words_with_weights.append({"word": word, "weight": weight})
feedback_context = json.dumps(queued_words_with_weights, indent=2)
full_prompt = f"{full_prompt}\n\nQueued feedback themes (with user-adjusted weights):\n{feedback_context}"
# Add historic feedback words if available (these may have weights too)
if historic_feedback_words:
# Extract just the words for historic context
historic_words = []
for item in historic_feedback_words:
key = list(item.keys())[0]
word = item[key]
historic_words.append(word)
feedback_historic_context = json.dumps(historic_words, indent=2)
full_prompt = f"{full_prompt}\n\nHistoric feedback themes (just words):\n{feedback_historic_context}"
return full_prompt
def _parse_feedback_response(self, response_content: str) -> List[str]:
"""Parse AI response to extract theme words."""
cleaned_content = self._clean_ai_response(response_content)
try:
data = json.loads(cleaned_content)
if isinstance(data, list):
theme_words = []
for word in data:
if isinstance(word, str):
theme_words.append(word.lower().strip())
else:
theme_words.append(str(word).lower().strip())
return theme_words
else:
logger.warning(f"AI returned unexpected data type for feedback: {type(data)}")
return []
except json.JSONDecodeError:
logger.warning("AI feedback response is not valid JSON, attempting to extract theme words...")
return self._extract_theme_words_from_text(response_content)
def _extract_theme_words_from_text(self, text: str) -> List[str]:
"""Extract theme words from plain text response."""
lines = text.strip().split('\n')
theme_words = []
for line in lines:
line = line.strip()
if line and len(line) < 50: # Theme words should be short
words = [w.lower().strip('.,;:!?()[]{}\"\'') for w in line.split()]
theme_words.extend(words)
if len(theme_words) >= 6:
break
return theme_words[:6]