app.py

import gradio as gr
import os
from llama_index import GPTSimpleVectorIndex
from langchain.agents import ZeroShotAgent, AgentExecutor
from langchain.agents import Tool
from langchain import OpenAI, LLMChain

os.environ['OPENAI_API_KEY'] = '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'

index = GPTSimpleVectorIndex.load_from_disk('/mnt/index/comboindex.json')


def querying_db(query: str):
    response = index.query(query)
    return response


tools = [
    Tool(
        name="QueryingDB",
        func=querying_db,
        description="useful for when you need to answer questions from the database. The answer is given in bullet points.",
        return_direct=True
    )
]

prefix = """Give a detailed answer to the question"""
suffix = """Give answer in bullet points"""

Question: {input}
{agent_scratchpad}"""

prompt = ZeroShotAgent.create_prompt(
    tools,
    prefix=prefix,
    suffix=suffix,
    input_variables=["input", "agent_scratchpad"]
)

llm_chain = LLMChain(llm=OpenAI(temperature=0), prompt=prompt)
tool_names = [tool.name for tool in tools]
agent = ZeroShotAgent(llm_chain=llm_chain, allowed_tools=tool_names)


def get_answer(query_string):
    agent_executor = AgentExecutor.from_agent_and_tools(agent=agent, tools=tools, verbose=True)
    result = agent_executor.run(query_string)

    return result


def qa_app(query):
    return get_answer(query)

inputs = gr.inputs.Textbox(label="Enter your question:")
output = gr.outputs.Textbox(label="Answer:")
iface = gr.Interface(fn=qa_app, inputs=inputs, outputs=output, title="Endo AI : Endocrine answering app by Dr. Om J Lakhani")

iface.launch()