The Journey to conversational interfaces
Download as PPTX, PDF3 likes618 views
This document summarizes the journey to building conversational interfaces using chatbots. It discusses starting with a help desk bot on Slack using NLP and machine learning via API.AI. Both rule-based and AI approaches are covered. The document outlines responding to questions, sending notifications, and analyzing data using the Slack API in Go. It also discusses notifying users, analyzing commands, and improving upon the initial architecture using general AI agents and NLP platforms. Key lessons learned include targeting multiple messaging platforms and choosing an ML engine.
![Slack Message Formats - 2
{
"attachments": [
{
"title": "New Ticket: Printer is not working",
"title_link": "https://api.slack.com/",
"text": "The Printer is not functioning….",
"fields": [
{
"title": "Priority",
"value": "High"
},
{
"title": "Reported by",
"value": "romin",
}
],
"footer": "My Help Desk system",
"ts": 1395676789
}
]
}](https://image.slidesharecdn.com/thejourneytoconversationalinterfaces-170225035330/85/The-Journey-to-conversational-interfaces-14-320.jpg)
The Journey to conversational interfaces
- 1. The Journey to Conversational Interfaces Romin Irani | Xoriant | @iRomin
- 2. Agenda Chatbot Ecosystem Our Story Help Desk Bots Initial Target Platform - Slack NLP + ML + API.AI
- 3. Chatbots! Chatbots! Chatbots! It is usually another way to interact with your Web Application, API or Mobile Application Multiple Messaging Platforms “Chatbots” are expected to be “The Thing” to do in 2017 ;-)
- 4. Rule-based v/s AI ● /find * contains=`xyz` ● /find pdf since=2d ● I’d like all documents that contain “xyz” ● Get me all PDF documents created in the last 2 days
- 5. Why Slack ? Popular Messaging Platform Integrates with Popular Applications APIs for Custom Integration Good Documentation App Store Starting point to learn more
- 6. Respond to Questions Notifications Customized Bot Analyze
- 7. Slack API - Go Language packages You can do everything with standard Go packages Alternately, here are a couple of packages to consider: https://github.com/nlopes/slack https://github.com/bluele/slack Skills needed Basic Go Syntax (Struct, Dealing with JSON)
- 10. Reporting Issue - Slash Command Slash Command → Messages that start with a slash (/) For e.g. /ticket or /report or /search /ticket listlist
- 11. Handling a Request from Slack http.HandleFunc("/ticket", MyTicketHandler) func MyTicketHandler(w http.ResponseWriter, r *http.Request) { //Extract out multiple params from Request //Validate if the request is coming from Slack and your team //Perform business logic //Compose the Response Struct json.NewEncoder(w).Encode(/* Slack Response Struct */)
- 12. Slack Incoming POST parameters token=oyIQ0AjzcBNgxGMcIsEumUpQ team_id=T0001 team_domain=example channel_id=C2147483705 channel_name=general user_id=U2147483697 user_name=romin command=/ticket text=list response_url=https://hooks.slack.com/commands/1234/5678
- 13. Slack Message Formats - 1 { "username": "Server Monitor" "text": "Cloud Storage Service is facing intermittent issues.nhttp://status.cloud.google.com", }
- 14. Slack Message Formats - 2 { "attachments": [ { "title": "New Ticket: Printer is not working", "title_link": "https://api.slack.com/", "text": "The Printer is not functioning….", "fields": [ { "title": "Priority", "value": "High" }, { "title": "Reported by", "value": "romin", } ], "footer": "My Help Desk system", "ts": 1395676789 } ] }
- 15. Notify
- 16. Sending Notifications - Incoming Webhook A mechanism to post a message to a Channel or a specific user New Tickets are routed as Notifications to respective Channels (Teams) Updates to Tickets are routed to the specific Users v := url.Values{"payload": {string(b)}} http.PostForm(WebhookURL, v)
- 17. Analyze
- 18. Analytics - Outgoing Webhook Calls your Webhook based on messages sent in a Channel Any message or specific Trigger Words Track Command Usage , Sentiment Analysis and more
- 19. Explosion of Commands … /ticket list /ticket create <title> <details> <priority> <department> /ticket update <status> <comment> /ticket close <id> /ticket view <id> /ticket help ….
- 20. Can we do better?
- 21. General AI Agents Messaging Platforms NLP , AI as a Service Platforms Frameworks & Deployment Platforms Source: https://www.oreilly.com/ideas/infographic-the-bot-platform-ecosystem
- 22. Our current architecture Messaging Platforms NLP Integrations Machine Learning Domains Bot Endpoint + Logic
- 23. Diagram inspired from : https://rasa.ai/ Which tickets are assigned to me? User input in Natural Language 1 NLP Tool 2 intent : tickets_search, entities: status : open, user : romin, Max_count : 5, Structured Data 3 Here are your tickets: 1…. 2… 3…. Bot output in Natural Language 5 How it works? Tickets API / Database 4
- 24. Lessons Learnt Go + Slack is a powerful combination to get going with your Bots. Need to target most popular Chat platforms. Build v/s Buying a Machine Learning Engine. Good Alternatives out there but choose carefully. Voice Interfaces. It is a journey.
- 25. Thank You
Editor's Notes
- #3: Help Desk System Deal with Tickets Legacy Web Application - PHP Based Research Bots and build out the Help Desk Chatbot Select a Platform to Host it Build in multiple other features like Notification & Analytics
- #5: Rule based Single purpose Understand a fixed set of commands Easier to build / host Only as smart as you program it to be AI based It has a brain NLP Ability to learn over time (Train) Complex to build out and host
- #8: Slash Command Incoming Webhook Outgoing Webhook Bot User Events API Realtime Message API (Websocket)
- #20: The Programmer in us took over We designed detailed command line syntax for multiple resources /report <title> <details> <priority> <department> /list <userid> /close <ticketid> Difficult to remember. /help ended up as the most used command!!! Could have created a conversation but it is a tough problem to address. Too constrained How can we make it feel more natural? NLP , AI , etc. Build v/s Buy What about other Chat Platforms?