app.py

import streamlit as st
import faiss
from transformers import pipeline
from sentence_transformers import SentenceTransformer
import json


def load_index():
    index = faiss.read_index('cdc_search.index')
    return index

def load_data():
    with open('./data.json') as f:
      data = json.load(f)
    return data


def load_embedder():
    embedder = SentenceTransformer("distilbert-base-nli-stsb-mean-tokens")
    return embedder


def load_qa_pipeline():
    qa = pipeline("question-answering", model="ktrapeznikov/albert-xlarge-v2-squad-v2")
    return qa


def load_questions():
    with open('./questions.json') as f:
      data = json.load(f)
    return (q for q in data)

index = load_index()
embedder = load_embedder()
qa = load_qa_pipeline()
data = load_data()

def search(query: str, k=1):
    encoded_query = embedder.encode([query])
    top_k = index.search(encoded_query, k)
    scores = top_k[0][0]
    results = [data[_id] for _id in top_k[1][0]]
    answers = []
    for result in results:
        answer = qa(question=query, context=result['text'])
        if 'answer' in answer:
            answers.append((answer['answer'], answer['score']))
    return sorted(answers, key=lambda tup: tup[1], reverse=True)

questions = load_questions()

option = st.selectbox("Sample Questions", questions)

st.write('You selected: ', option)

st.markdown("\n".join([f"* {answer}" for (answer, _) in search(option)]))