Lex

Build conversational chatbots using the same technology powering Alexa

Amazon Lex is a fully managed service for building conversational interfaces - chatbots and voice assistants - using the same deep learning technology that powers Alexa. It handles automatic speech recognition (ASR) for voice and natural language understanding (NLU) for text, letting you define intents, utterances, and slot types to build sophisticated multi-turn conversations. For cloud engineers, Lex is the backbone of customer service automation, IVR replacement, and internal productivity bots.

Lex Core Concepts - Intents, Slots, and Utterances

Understanding Lex's data model is essential before building any bot. Everything in Lex is built around these four concepts.

Concept	Definition	Example
Intent	An action the user wants to perform	BookFlight, CheckOrderStatus, ResetPassword
Utterance	Sample phrases that map to an intent (you provide 10-20)	"I want to fly to London", "Book me a flight", "Get me a ticket to NYC"
Slot	A variable that needs to be filled to complete the intent	DepartureCity, ArrivalCity, TravelDate, CabinClass
Slot Type	The data type and valid values for a slot	Custom list (Economy, Business, First), Built-in (AMAZON.City, AMAZON.Date)
Fulfillment	What happens after all slots are filled	Lambda function call or return control to calling app
Elicitation	Bot prompts to collect missing slot values	"What city are you flying to?" - asked until DepartureCity is filled

💡

Lex V2 (current version) uses a concept called "conversation flows" with visual builder support. Lex V1 is legacy - all new bots should use V2. V2 also supports multi-language bots in a single bot definition.

Lambda Integration - Fulfillment and Validation Hooks

Lambda functions are the primary way to add business logic to Lex bots. There are two hook points: the dialog code hook (called after each user turn for validation) and the fulfillment code hook (called after all slots are filled).

bash

# Lambda fulfillment handler for Lex V2
import json

def lambda_handler(event, context):
    intent_name = event['sessionState']['intent']['name']
    slots = event['sessionState']['intent']['slots']
    
    if intent_name == 'BookFlight':
        departure = slots['DepartureCity']['value']['interpretedValue']
        arrival = slots['ArrivalCity']['value']['interpretedValue']
        travel_date = slots['TravelDate']['value']['interpretedValue']
        
        # Business logic here (call booking API, check availability, etc.)
        booking_ref = book_flight(departure, arrival, travel_date)
        
        return {
            'sessionState': {
                'dialogAction': {'type': 'Close'},
                'intent': {'name': intent_name, 'state': 'Fulfilled'}
            },
            'messages': [{
                'contentType': 'PlainText',
                'content': f'Your flight from {departure} to {arrival} on {travel_date} is booked. Ref: {booking_ref}'
            }]
        }

Hook Type	When Called	Can Do	Return
Dialog Code Hook	After every user turn	Validate slots, re-elicit, add context	Delegate, ElicitSlot, ElicitIntent, Close
Fulfillment Code Hook	After all slots filled	Execute business logic, call APIs	Close (success or failure message)

Deployment Channels and Integration Patterns

Lex bots can be deployed to multiple channels without changing the bot definition. AWS provides native integrations for the most common contact center and messaging platforms.

Channel	Integration Method	Use Case
Amazon Connect	Native Lex integration in contact flows	IVR replacement, voice + chat customer service
Facebook Messenger	Lex console channel configuration	Consumer-facing chatbot on Facebook pages
Slack	Lex console channel configuration	Internal IT helpdesk, DevOps bots
Twilio SMS	Lex console channel configuration	SMS-based customer support
Web/Mobile App	AWS Amplify AI or direct API calls	Custom UI with Lex backend
Genesys / Salesforce	Partner integrations	Enterprise contact center platforms

💡

Amazon Connect + Lex is the most common enterprise pattern. You can build a full IVR replacement where Lex handles natural language intents, and Connect routes to the right agent queue based on the Lex session attributes.

Lex Pricing

Feature	Pricing	Notes
Text requests	$0.00075 per request	One request = one user message turn
Speech requests	$0.004 per request	Voice input uses both ASR and NLU
Free tier	10,000 text + 5,000 speech requests/month	Free for 12 months

For Amazon Connect integration, Lex charges are in addition to Connect charges. At scale, a contact center handling 1 million voice calls per month at 5 turns per call would incur $20,000/month in Lex charges alone - factor this into pricing models.

🎯

Interview Focus Points

1Explain the Lex data model - what are intents, slots, utterances, and how do they relate to each other?
2What is the difference between a dialog code hook and a fulfillment code hook in Lex?
3How would you build an IVR system using Amazon Lex and Amazon Connect?
4How does Lex handle ambiguous user input where the utterance could match multiple intents?
5How would you build a multi-language customer support bot using Lex V2?
6What is the slot elicitation process and how do you implement validation logic in Lambda?
7How would you test and iterate on a Lex bot in a CI/CD pipeline?