IntelliBot: AI-Powered Q&A Solution for Enhanced Customer Support

Problem Statement

As part of this project, we built an AI driven chatbot that makes use of a Large Language Model (LLM), a vector database, and sophisticated embedding techniques. It designed to improve the caliber and efficiency of customer care by offering users of telecom routers and switching equipment exact, real-time assistance.

Tech Stacks

Python

LLM

Word Embedding

Fine Tuning

Vector DB

Flask

GIT

Pickle

Technologies listed above were utilized to implement the solutions for this project.

Sources of Data

Internal Data

Databases

Explain the Dataset

This project is based on multiple PDF files that has in depths information about the products and an excel sheet regarding one of the devices with two types of problems solutions. We are now leveraging this data to build a responsive solutions model in the form of chatbot, using Large Language Model(LLM) and advanced embeddings unrelated. For this, the chatbot is designed in a way to help users of telecom routers and switching equipment as suggested by them which ensures that customer care experience goes smooth and positive.

Workflow Architecture

These steps are involved in developing and implementing this project are described in the architecture below.

Project Architecture for Propensity Model

Sample Code Snippets

    
# app.py
from flask import Flask, request, jsonify
import chromadb
from chromadb.config import Settings
import torch
from transformers import AutoModelForSequenceClassification
from sentence_transformers import SentenceTransformer

app = Flask(__name__)

# Load model, tokenizer, and embedding model
model = AutoModelForSequenceClassification.from_pretrained('./saved_model')
embedding_model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')

# Initialize Chroma DB
chroma_client = chromadb.Client(Settings(chroma_api_impl='chromadb.impl.sqlite.SQLiteAPI'))
collection = chroma_client.get_collection('telecom_support')

def get_prediction(text):
    # Encode the text using the embedding model
    embedding = embedding_model.encode(text, convert_to_tensor=True).numpy()
    
    # Query Chroma DB
    results = collection.query(embedding=embedding)
    
    # Retrieve the top result
    top_result = results[0]
    
    # Predict using the Falcon model
    inputs = torch.tensor([embedding])
    outputs = model(inputs)
    prediction = torch.argmax(outputs.logits, dim=1)
    
    return top_result['metadata']['label']

@app.route('/predict', methods=['POST'])
def predict():
    data = request.get_json(force=True)
    text = data['text']
    prediction = get_prediction(text)
    return jsonify({'prediction': prediction})

if __name__ == '__main__':
    app.run(debug=True)