Building NLP-Powered Chatbots: From Rules to Transformers

Evolution of Approaches:

• Rule-based: Pattern matching (ELIZA, AIML)
• Intent-based: NLU + Dialog management (Rasa, Dialogflow)
• Generative: Transformer models (GPT, BlenderBot)
• Hybrid: Combines multiple approaches

Architecture Comparison:

Key Components:

Rasa Implementation:

# nlu.yml example
version: "3.1"
nlu:
- intent: greet
  examples: |
    - hello
    - hi there
    - good morning
    
# domain.yml
intents:
  - greet
responses:
  utter_greet:
  - text: "Hello! How can I help?"

# stories.yml
stories:
- story: greet path
  steps:
  - intent: greet
  - action: utter_greet
        

Training Pipeline:

from rasa.train import train_nlu
train_nlu(
    config="config.yml",
    nlu_data="nlu.yml",
    output="models/"
)

from rasa.core.agent import Agent
agent = Agent.load("models/")
agent.handle_text("hello")  # Returns: "Hello! How can I help?"
        

Key Features:

• End-to-end: Single model handles conversation
• Context-aware: Maintains conversation history
• Multiturn: Handles complex dialogs
• Pretrained: Leverages models like DialoGPT

HuggingFace Implementation:

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

tokenizer = AutoTokenizer.from_pretrained("microsoft/DialoGPT-medium")
model = AutoModelForCausalLM.from_pretrained("microsoft/DialoGPT-medium")

# Chat loop
chat_history_ids = None
while True:
    user_input = input(">> User: ")
    if user_input.lower() == 'quit':
        break
        
    input_ids = tokenizer.encode(user_input + tokenizer.eos_token, 
                               return_tensors='pt')
    bot_input_ids = torch.cat([chat_history_ids, input_ids], dim=-1) if chat_history_ids is not None else input_ids
    
    chat_history_ids = model.generate(
        bot_input_ids,
        max_length=1000,
        pad_token_id=tokenizer.eos_token_id,
        temperature=0.7,
        top_k=50,
        top_p=0.9
    )
    
    print("Bot:", tokenizer.decode(chat_history_ids[:, bot_input_ids.shape[-1]:][0], 
          skip_special_tokens=True))
        

Prompt Engineering Tips:

"""You are a helpful travel assistant. 
Provide concise, accurate information about flights and hotels.
User: What's the weather in Paris?"""
            

"""Q: What's the capital of France? 
A: Paris
Q: What's the population?
A: About 2.1 million
Q: What's the weather like?"""
            

Production Considerations:

• Channels: Web, Mobile, WhatsApp, etc.
• Orchestration: Message routing
• Monitoring: Conversation analytics
• Fallbacks: Handoff to human agents

Optimization Techniques:

Deployment Example:

# FastAPI endpoint for chatbot
from fastapi import FastAPI
from pydantic import BaseModel

app = FastAPI()

class ChatInput(BaseModel):
    message: str
    conversation_id: str = None

@app.post("/chat")
async def chat_endpoint(input: ChatInput):
    response = agent.handle_message(input.message)
    return {
        "response": response,
        "conversation_id": input.conversation_id
    }

# Deploy with: uvicorn main:app --host 0.0.0.0 --port 8000