Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| #import openai | |
| import faiss | |
| import numpy as np | |
| import json | |
| import os | |
| from openai import OpenAI | |
| from openai import OpenAI | |
| client = OpenAI(api_key=os.getenv("OPENAI_API_KEY")) | |
| # Load the FAQs with embeddings | |
| with open('faqs_with_embeddings.json', 'r') as f: | |
| faqs_with_embeddings = json.load(f) | |
| # Load the FAISS index | |
| index = faiss.read_index('faiss_index.index') | |
| def get_relevant_faqs(query, index, faqs_with_embeddings, client): | |
| query_embedding_response = client.embeddings.create( | |
| input=query, model="text-embedding-3-large" | |
| ) | |
| query_embedding = query_embedding_response.data[0].embedding | |
| query_embedding = np.array(query_embedding).astype('float32').reshape(1, -1) | |
| distances, indices = index.search(query_embedding, k=15) # Retrieve top 7 matches | |
| results = [] | |
| for i in indices[0]: | |
| results.append(faqs_with_embeddings[i]) | |
| return results | |
| def create_prompt(user_query, relevant_faqs): | |
| system_prompt = ( | |
| "You are a customer service agent for a bank named HBL Microfinance Bank that has a mobile wallet app named FirstPay. You are required to answer the customer query given below by looking at the information from FAQs list given after the customer query. If the information in the FAQs list is not sufficient to answer the customer query then inform the customer to call the helpline. Where needed you can also augment the answers using your general knowledge, but make sure that information about FirstPay Wallet and HBL Microfinance is coming fom the FAQs list provided and do not make anything yourself about those. Customers are from Pakistan and the customer query may also be regional Pakistani languages such as Urdu, Pashto, Sindhi, Saraiki etc and the fonts may be either roman fonts or some other regional languages fonts, if the customer asks question in one of the regional languages then answer them in the same regional language used by the customer but use roman english fonts or alphabets only." | |
| #"You are a helpful assistant. Use the provided FAQs to answer the user's query. " | |
| #"If the FAQs do not contain enough information, use your general knowledge to help." | |
| ) | |
| faqs_context = "\n\n".join( | |
| [f"FAQ {i+1}:\nQ: {faq['question']}\nA: {faq['answer']}" for i, faq in enumerate(relevant_faqs)] | |
| ) | |
| user_prompt = f"User query: {user_query}\n\nFAQs:\n{faqs_context}" | |
| return system_prompt, user_prompt | |
| def chatbot_response(query): | |
| relevant_faqs = get_relevant_faqs(query, index, faqs_with_embeddings, client) | |
| system_prompt, user_prompt = create_prompt(query, relevant_faqs) | |
| response = client.chat.completions.create( | |
| model="gpt-4o", # You can use "gpt-4" if you have access | |
| messages=[ | |
| {"role": "system", "content": system_prompt}, | |
| {"role": "user", "content": user_prompt} | |
| ], | |
| temperature=0.0 | |
| ) | |
| #answer = response['choices'][0]['message']['content'] | |
| answer = response.choices[0].message.content | |
| return answer | |
| # Create the Gradio interface | |
| iface = gr.Interface( | |
| fn=chatbot_response, | |
| inputs="text", | |
| outputs="text", | |
| title="FirstPay Demo FAQ Chatbot", | |
| description="I am your (Demo) AI Assistant for FirstPay, how can I help you." | |
| ) | |
| # Launch the interface | |
| iface.launch(share=True) | |