Pattern-Matching Chatbot (No API Needed)

• Build keyword/regex intent matching.
• Map intents to scripted responses (with variety).
• Add a small "memory" (the user's name, last topic).
• Understand the limits — and why LLMs (next lessons) are different.

User: I am feeling sad
Bot:  Why do you say you are feeling sad?

User: my homework is hard
Bot:  Tell me more about your homework.

ELIZA, the first famous chatbot, used pure pattern-matching — reflecting your words back. No understanding at all, yet people felt heard. That's the power (and danger) of scripted bots.

Intent matching with keywords

import re, random

INTENTS = {
    "greeting": (["hello", "hi", "hey", "yo"],
                 ["Hi there!", "Hello! How can I help?"]),
    "bye":      (["bye", "goodbye", "see you"],
                 ["Goodbye!", "See you later!"]),
    "hours":    (["open", "hours", "time"],
                 ["We're open 9am-6pm, Mon-Sat."]),
    "thanks":   (["thanks", "thank you"],
                 ["You're welcome!", "Anytime!"]),
}

def detect_intent(text):
    text = text.lower()
    for intent, (keywords, _) in INTENTS.items():
        if any(re.search(rf"\b{k}\b", text) for k in keywords):
            return intent
    return None

Respond

def respond(text):
    intent = detect_intent(text)
    if intent:
        return random.choice(INTENTS[intent][1])
    return "Sorry, I didn't understand. Try asking about our hours."

The chat loop

while True:
    user = input("You: ")
    if user.lower() in ("quit", "exit"):
        break
    print("Bot:", respond(user))

Regex for richer patterns

# capture a name: "my name is Aisyah"
m = re.search(r"my name is (\w+)", text.lower())
if m:
    name = m.group(1).title()
    return f"Nice to meet you, {name}!"

The hard limit

Add a rule for every phrasing and you'll never finish — language has infinite variety. Rule bots break the moment a user phrases something unexpectedly. That brittleness is exactly why neural LLMs took over open-ended conversation.

# cafe_bot.py — rule-based FAQ bot with a little memory
import re, random

INTENTS = {
    "greeting": (["hello", "hi", "hey"],
                 ["Hi! Welcome to Kopi Corner ☕", "Hello there!"]),
    "menu":     (["menu", "drinks", "coffee", "food"],
                 ["We serve kopi, teh tarik, toast and nasi lemak."]),
    "hours":    (["open", "hours", "close", "time"],
                 ["Open daily 7am-7pm."]),
    "location": (["where", "located", "address"],
                 ["We're at 12 Jalan Mawar, KL."]),
    "thanks":   (["thanks", "thank"],
                 ["You're welcome! 😊"]),
    "bye":      (["bye", "goodbye"],
                 ["Jumpa lagi! 👋"]),
}

memory = {}

def respond(text):
    low = text.lower()
    # memory: capture name
    m = re.search(r"my name is (\w+)", low)
    if m:
        memory["name"] = m.group(1).title()
        return f"Nice to meet you, {memory['name']}!"

    for intent, (keys, replies) in INTENTS.items():
        if any(re.search(rf"\b{k}\b", low) for k in keys):
            reply = random.choice(replies)
            if "name" in memory and intent == "greeting":
                reply = f"{reply} Good to see you again, {memory['name']}!"
            return reply
    return "I'm just a simple bot — try asking about the menu, hours, or location."

if __name__ == "__main__":
    print("Bot: Hi! Ask me about Kopi Corner. (type 'bye' to leave)")
    while True:
        user = input("You: ").strip()
        reply = respond(user)
        print("Bot:", reply)
        if "bye" in user.lower():
            break

Sample chat

You: hi, my name is Aisyah
Bot: Nice to meet you, Aisyah!
You: what's on the menu?
Bot: We serve kopi, teh tarik, toast and nasi lemak.
You: when do you close?
Bot: Open daily 7am-7pm.
You: tell me a joke
Bot: I'm just a simple bot — try asking about the menu, hours, or location.

Read the diff

Intents + scripted replies + a tiny memory dict = a usable FAQ bot, no API, no cost. But watch the last line: "tell me a joke" falls through to the fallback. A rule bot can only do what you scripted. Tomorrow we connect a real LLM that handles anything.

A rule bot detects an intent (keywords/regex/TF-IDF) and returns a scripted reply, optionally with a little memory. It's free, fast, and reliable for narrow domains — and brittle outside them. That brittleness is the whole reason LLMs exist. From the next lesson, we let a real language model handle open-ended conversation.