dewey-oracle/app/recommender.py

import google.generativeai as genai
from typing import List, Dict, Any
import json
from app.config import get_settings


class BookRecommender:
    """AI-powered book recommendation engine using Gemini."""

    def __init__(self):
        settings = get_settings()
        if not settings.gemini_api_key:
            raise ValueError("GEMINI_API_KEY not configured")
        genai.configure(api_key=settings.gemini_api_key)
        # Use the latest flash model available
        self.model = genai.GenerativeModel('models/gemini-2.5-flash')

    async def generate_recommendations(
        self,
        reading_history: List[Dict[str, Any]],
        num_recommendations: int = 5
    ) -> List[Dict[str, Any]]:
        """
        Generate book recommendations based on reading history.

        Args:
            reading_history: List of books the user has read/listened to
            num_recommendations: Number of recommendations to generate

        Returns:
            List of recommended books with title, author, description, and reason
        """
        # Build context from reading history
        history_text = self._format_reading_history(reading_history)

        # Create prompt for Gemini
        prompt = f"""Based on this reading history, recommend {num_recommendations} audiobooks that this person would enjoy.

Reading History:
{history_text}

For each recommendation, provide:
1. Title
2. Author
3. Brief description (2-3 sentences)
4. Why you're recommending it based on their reading history (1-2 sentences)
5. Genres (as a list)

Format your response as a JSON array with objects containing: title, author, description, reason, genres.
Only respond with the JSON array, no additional text."""

        # Call Gemini API
        response = self.model.generate_content(prompt)

        # Parse response
        response_text = response.text
        try:
            recommendations = json.loads(response_text)
            return recommendations
        except json.JSONDecodeError:
            # If Claude didn't return valid JSON, try to extract it
            # Look for JSON array in the response
            start = response_text.find("[")
            end = response_text.rfind("]") + 1
            if start >= 0 and end > start:
                recommendations = json.loads(response_text[start:end])
                return recommendations
            else:
                raise ValueError("Failed to parse recommendations from AI response")

    def _format_reading_history(self, history: List[Dict[str, Any]]) -> str:
        """Format reading history for the AI prompt."""
        formatted = []
        for i, book in enumerate(history, 1):
            title = book.get("title", "Unknown")
            author = book.get("author", "Unknown")
            genres = book.get("genres", [])
            progress = book.get("progress", 0)
            is_finished = book.get("is_finished", False)

            status = "Finished" if is_finished else f"In Progress ({int(progress * 100)}%)"
            genre_str = ", ".join(genres) if genres else "Unknown"

            formatted.append(
                f"{i}. {title} by {author}\n"
                f"   Status: {status}\n"
                f"   Genres: {genre_str}"
            )

        return "\n\n".join(formatted)

    async def explain_recommendation(
        self,
        book_title: str,
        book_author: str,
        reading_history: List[Dict[str, Any]]
    ) -> str:
        """
        Get a detailed explanation for why a specific book is recommended.

        Args:
            book_title: Title of the recommended book
            book_author: Author of the recommended book
            reading_history: User's reading history

        Returns:
            Detailed explanation text
        """
        history_text = self._format_reading_history(reading_history)

        prompt = f"""Explain in detail why "{book_title}" by {book_author} would be a good recommendation for someone with this reading history:

{history_text}

Provide a thoughtful 2-3 paragraph explanation connecting specific aspects of the recommended book to their reading preferences."""

        response = self.model.generate_content(prompt)

        return response.text