src/rag.py

import os
import json
import re

DATA_DIR = os.path.dirname(os.path.abspath(__file__)) + "/data"
GUIDES_FILE = DATA_DIR + "/agri_guides.json"

def retrieve_guides(query_text, crop_name, lang="hi"):
    """
    Search agri_guides.json for guides relevant to the query and crop.
    Returns (markdown_grounding_text, source_list) or (None, [])
    """
    if not os.path.exists(GUIDES_FILE):
        print(f"[rag.py] Guide file not found at {GUIDES_FILE}")
        return None, []

    try:
        with open(GUIDES_FILE, "r", encoding="utf-8") as f:
            guides = json.load(f)
    except Exception as e:
        print(f"[rag.py] Error reading guides: {e}")
        return None, []

    # Clean query text
    query_text_lower = query_text.lower()
    query_words = set(re.findall(r"\w+", query_text_lower))
    
    matches = []
    
    for guide in guides:
        # Filter by crop (case-insensitive check)
        if guide["crop"].lower() != crop_name.lower():
            continue
            
        score = 0
        # Check keyword matches
        for keyword in guide.get("keywords", []):
            if keyword.lower() in query_text_lower:
                score += 3  # High weight for exact keyword matching
                
        # Check general word overlap in query
        guide_text = (guide.get("title_en", "") + " " + guide.get("content_en", "") + " " +
                      guide.get("title_hi", "") + " " + guide.get("content_hi", "")).lower()
        
        for word in query_words:
            if len(word) > 2 and word in guide_text:
                score += 1
                
        if score > 0:
            matches.append((score, guide))
            
    # Sort matches by score descending
    matches.sort(key=lambda x: x[0], reverse=True)
    
    if not matches:
        return None, []
        
    # Build markdown grounding context and source citation list
    grounding_parts = []
    sources = []
    
    # Take top-2 matching guides
    for idx, (score, guide) in enumerate(matches[:2]):
        title = guide["title_en"] if lang in ["en", "hinglish"] else guide["title_hi"]
        content = guide["content_en"] if lang in ["en", "hinglish"] else guide["content_hi"]
        topic = guide["topic"]
        
        grounding_parts.append(
            f"### {title}\n"
            f"{content}\n"
        )
        sources.append(f"{guide['crop']} Guide: {topic}")
        
    grounding_text = "\n---\n".join(grounding_parts)
    return grounding_text, sources