nlu-core 1.0.0

Multilingual NLU engine for intent classification, escalation detection, and conversation management.

nlu-core

Multilingual Natural Language Understanding (NLU) engine for customer support chatbots.

Supports English and Bahasa Malaysia (Malay).

Features

• Intent classification — maps customer messages to knowledge-base topics
• Escalation detection — identifies customers who need a human agent
• Conversation summarisation — produces compact conversation summaries for context
• History reply — finds the most relevant previous answer in a conversation
• Product enquiry prompts — builds structured prompts for product recommendation flows

Installation

pip install nlu-core

Quick Start

import pandas as pd
from nlu_core import engine_match, detect_escalation, summarize_conversation

# Load your knowledge base
kb = pd.read_excel("knowledge_base.xlsx", sheet_name="Main DB")

# Classify a customer message
intent, confidence, matched_row = engine_match(
    user_question="How do I track my order?",
    knowledge_df=kb,
)

# Check for escalation
should_escalate, response = detect_escalation("I want to speak to a manager!")

# Summarise a conversation
summary = summarize_conversation(
    history=["Customer: My phone is broken.", "Agent: I can help with that."],
)

API Reference

Function	Description
engine_match(user_question, knowledge_df, ...)	Returns (intent, confidence, matched_row)
detect_escalation(question)	Returns (should_escalate: bool, response: str)
summarize_conversation(history, ...)	Returns a summary string
find_relevant_history_reply(history, question, ...)	Returns the best matching previous reply
build_product_enquiry_prompt(user_message, stock_json)	Returns a structured LLM prompt

Constants

Name	Description
DEFAULT_GEMINI_MODEL	Default primary language model
DEFAULT_OPENAI_MODEL	Default translation model
FALLBACK_GEMINI_MODEL	Fallback model if primary is unavailable
LOG_TICKET	Intent token for support ticket creation
MATCH_GEMINI_MODEL	Intent token for product matching

Requirements

• Python 3.10+
• macOS (Apple Silicon or Intel)
• google-genai
• openai
• pandas