Reklamator/app/services/ai_analyzer.py

"""AI-powered feedback analysis service"""
from abc import ABC, abstractmethod
import anthropic
import re
from app.models.feedback import AnalysisResult


class AIAnalyzer(ABC):
    """Abstract base class for AI-powered feedback analyzers

    Subclasses must implement the analyze_feedback method to provide
    categorization, summarization, and translation capabilities.
    """

    @abstractmethod
    def analyze_feedback(self, feedback_text, target_language, product_id):
        """Analyze feedback using AI

        Args:
            feedback_text: The feedback text to analyze
            target_language: Language code for translation (e.g., 'en', 'de')
            product_id: Product ID for context

        Returns:
            AnalysisResult: Analysis results including category, summary, and translation
        """
        pass


class ClaudeAnalyzer(AIAnalyzer):
    """Claude AI-based feedback analyzer using Anthropic API

    Uses Claude to analyze feedback and extract:
    - Category (bug, feature_request, question, complaint, praise, other)
    - Original language detection
    - Summary (1-2 sentences)
    - Translation to target language
    """

    # Valid feedback categories
    VALID_CATEGORIES = ['bug', 'feature_request', 'question', 'complaint', 'praise', 'other']

    def __init__(self, api_key):
        """Initialize Claude analyzer

        Args:
            api_key: Anthropic API key
        """
        self.api_key = api_key
        self.client = anthropic.Anthropic(api_key=api_key)

    def analyze_feedback(self, feedback_text, target_language, product_id):
        """Analyze feedback using Claude API

        Args:
            feedback_text: The feedback text to analyze
            target_language: Language code for translation (e.g., 'en', 'de')
            product_id: Product ID for context

        Returns:
            AnalysisResult: Analysis results

        Raises:
            Exception: If API call fails or timeout occurs
        """
        # Design prompt for Claude API (T075)
        prompt = self._build_analysis_prompt(feedback_text, target_language)

        try:
            # Call Claude API with timeout (T074, T080)
            response = self._call_claude_api(prompt, timeout=45)

            # Extract analysis components from response
            raw_analysis = response.content[0].text

            # Extract category (T077)
            category = self._extract_category(raw_analysis)

            # Detect original language (T076)
            original_language = self._extract_language(raw_analysis)

            # Extract summary (T078)
            summary = self._extract_summary(raw_analysis)

            # Extract translation (T079)
            translation = self._extract_translation(raw_analysis)

            return AnalysisResult(
                category=category,
                original_language=original_language,
                summary=summary,
                translation=translation,
                raw_analysis=raw_analysis
            )

        except anthropic.APITimeoutError as e:
            # Handle API timeouts (T080)
            raise Exception(f"Claude API timeout after 45s: {str(e)}")
        except anthropic.APIError as e:
            # Handle API errors with retry logic (T081)
            raise Exception(f"Claude API error: {str(e)}")
        except Exception as e:
            # General error handling
            raise Exception(f"Analysis failed: {str(e)}")

    def _build_analysis_prompt(self, feedback_text, target_language):
        """Build the analysis prompt for Claude

        Prompt design (T075): Single call to categorize, summarize, and translate
        """
        return f"""Analyze the following user feedback and provide a structured analysis.

User Feedback:
{feedback_text}

Please provide your analysis in the following format:

# Feedback Analysis

**Category**: [Choose ONE: bug, feature_request, question, complaint, praise, other]

**Original Language**: [Detect the language code, e.g., en, de, fr, es]

**Summary**: [Provide a concise 1-2 sentence summary of the feedback]

**Translation**: [Translate the feedback to {target_language}. If already in {target_language}, write "(same as original)"]

Important:
- Be accurate in language detection
- Choose the most appropriate category
- Keep the summary brief but informative
- Translate naturally and accurately"""

    def _call_claude_api(self, prompt, timeout=45):
        """Call Claude API with proper configuration

        Args:
            prompt: The prompt to send to Claude
            timeout: Timeout in seconds (default 45s per T080)

        Returns:
            API response object

        Raises:
            anthropic.APITimeoutError: If request times out
            anthropic.APIError: If API returns an error
        """
        return self.client.messages.create(
            model="claude-3-5-sonnet-20241022",
            max_tokens=1000,
            timeout=timeout,
            messages=[
                {
                    "role": "user",
                    "content": prompt
                }
            ]
        )

    def _extract_category(self, analysis_text):
        """Extract category from analysis text (T077)

        Args:
            analysis_text: Raw analysis markdown

        Returns:
            str: Category (defaults to 'other' if not found or invalid)
        """
        # Look for pattern: **Category**: bug
        match = re.search(r'\*\*Category\*\*:\s*(\w+)', analysis_text, re.IGNORECASE)

        if match:
            category = match.group(1).lower()
            # Validate category
            if category in self.VALID_CATEGORIES:
                return category

        # Default to 'other' if not found or invalid
        return 'other'

    def _extract_language(self, analysis_text):
        """Extract detected language from analysis text (T076)

        Args:
            analysis_text: Raw analysis markdown

        Returns:
            str: Language code (defaults to 'unknown' if not found)
        """
        # Look for pattern: **Original Language**: en
        match = re.search(r'\*\*Original Language\*\*:\s*(\w+)', analysis_text, re.IGNORECASE)

        if match:
            return match.group(1).lower()

        # Default to 'unknown'
        return 'unknown'

    def _extract_summary(self, analysis_text):
        """Extract summary from analysis text (T078)

        Args:
            analysis_text: Raw analysis markdown

        Returns:
            str: Summary text (defaults to empty string if not found)
        """
        # Look for pattern: **Summary**: [text]
        match = re.search(r'\*\*Summary\*\*:\s*(.+?)(?=\n\*\*|\n\n|$)', analysis_text, re.IGNORECASE | re.DOTALL)

        if match:
            return match.group(1).strip()

        # Default to empty string
        return ''

    def _extract_translation(self, analysis_text):
        """Extract translation from analysis text (T079)

        Args:
            analysis_text: Raw analysis markdown

        Returns:
            str: Translated text (defaults to empty string if not found)
        """
        # Look for pattern: **Translation**: [text]
        match = re.search(r'\*\*Translation\*\*:\s*(.+?)(?=\n\*\*|\n\n|$)', analysis_text, re.IGNORECASE | re.DOTALL)

        if match:
            return match.group(1).strip()

        # Default to empty string
        return ''