# query.py
#
# Run a query on a vector store
#
# August 2025
# E. M. Furst

from llama_index.core import (
    load_index_from_storage,
    StorageContext,
    Settings,
)
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.llms.ollama import Ollama
from llama_index.core.prompts import PromptTemplate
import os, time

#
# Globals
#
os.environ["TOKENIZERS_PARALLELISM"] = "false"

# Embedding model used in vector store (this should match the one in build.py)
embed_model = HuggingFaceEmbedding(cache_folder="./models",
    model_name="BAAI/bge-large-en-v1.5")

# LLM model to use in query transform and generation
llm = "command-r7b"

#
# Custom prompt for the query engine
#
PROMPT = PromptTemplate(
"""You are an expert research assistant. You are given top-ranked writing \
excerpts (CONTEXT) and a user's QUERY.

Instructions:
- Base your response *only* on the CONTEXT.
- The snippets are ordered from most to least relevant—prioritize insights \
from earlier (higher-ranked) snippets.
- Aim to reference *as many distinct* relevant files as possible (up to 10).
- Do not invent or generalize; refer to specific passages or facts only.
- If a passage only loosely matches, deprioritize it.

Format your answer in two parts:

1. **Summary Theme**
   Summarize the dominant theme from the relevant context in a few sentences.

2. **Matching Files**
   Make a list of 10 matching files. The format for each should be:
   <filename> - <rationale tied to content. Include date if available.>

CONTEXT:
{context_str}

QUERY:
{query_str}

Now provide the theme and list of matching files."""
)

#
# Main program routine
#

def main():
    # Use a local model to generate -- in this case using Ollama
    Settings.llm = Ollama(
        model=llm,
        request_timeout=360.0,
    )

    # Load embedding model (same as used for vector store)
    Settings.embed_model = embed_model

    # Load persisted vector store + metadata
    storage_context = StorageContext.from_defaults(persist_dir="./storage")
    index = load_index_from_storage(storage_context)

    # Build regular query engine with custom prompt
    query_engine = index.as_query_engine(
        similarity_top_k=15,
        text_qa_template=PROMPT,
    )

    # Query
    while True:
        q = input("\nEnter a search topic or question (or 'exit'): ").strip()
        if q.lower() in ("exit", "quit"):
            break
        print()

        # Generate the response by querying the engine
        start_time = time.time()
        response = query_engine.query(q)
        end_time = time.time()

        # Return the query response and source documents
        print(response.response)

        print("\nSource documents:")
        for node in response.source_nodes:
            meta = getattr(node, "metadata", None) or node.node.metadata
            print(f"  {meta.get('file_name')}  {getattr(node, 'score', None)}")

        print(f"\nElapsed time: {(end_time-start_time):.1f} seconds")

if __name__ == "__main__":
    main()