Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
We Make Bots. For Real
Search
Андрей Листочкин (Andrey Listochkin)
April 08, 2017
Programming
0
420
We Make Bots. For Real
Андрей Листочкин (Andrey Listochkin)
April 08, 2017
Tweet
Share
More Decks by Андрей Листочкин (Andrey Listochkin)
See All by Андрей Листочкин (Andrey Listochkin)
Everybody Stand Back! I Know Regular Expressions
listochkin
0
170
Command-line scripting with Rust. Wait, what?!
listochkin
0
350
Server Memory - BuildStuff Ukraine 2019
listochkin
0
43
Server Memory - Chernivtsi JS 2019
listochkin
1
150
10 Years Later
listochkin
0
360
Managing Managers - DevTalks iHUB
listochkin
0
56
Time, Numbers, Text
listochkin
1
590
Software Licensing: A Minefield Guide
listochkin
0
150
Devops for A Busy Developer - XP Days 2016
listochkin
0
180
Other Decks in Programming
See All in Programming
SQLアンチパターン第2版 データベースプログラミングで陥りがちな失敗とその対策 / Intro to SQL Antipatterns 2nd
twada
PRO
38
11k
Claude Code で Astro blog を Pages から Workers へ移行してみた
codehex
0
180
AI Ramen Fight
yusukebe
0
130
20250808_AIAgent勉強会_ClaudeCodeデータ分析の実運用〜競馬を題材に回収率100%の先を目指すメソッドとは〜
kkakeru
0
120
書き捨てではなく継続開発可能なコードをAIコーディングエージェントで書くために意識していること
shuyakinjo
1
250
Reactの歴史を振り返る
tutinoko
1
180
コーディングは技術者(エンジニア)の嗜みでして / Learning the System Development Mindset from Rock Lady
mackey0225
2
300
抽象化という思考のツール - 理解と活用 - / Abstraction-as-a-Tool-for-Thinking
shin1x1
1
950
No Install CMS戦略 〜 5年先を見据えたフロントエンド開発を考える / no_install_cms
rdlabo
0
480
Constant integer division faster than compiler-generated code
herumi
2
560
管你要 trace 什麼、bpftrace 用下去就對了 — COSCUP 2025
shunghsiyu
0
370
一人でAIプロダクトを作るための工夫 〜技術選定・開発プロセス編〜 / I want AI to work harder
rkaga
10
2.2k
Featured
See All Featured
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
PRO
22
1.4k
Why You Should Never Use an ORM
jnunemaker
PRO
58
9.5k
Git: the NoSQL Database
bkeepers
PRO
431
65k
Mobile First: as difficult as doing things right
swwweet
223
9.9k
Chrome DevTools: State of the Union 2024 - Debugging React & Beyond
addyosmani
7
800
Building Applications with DynamoDB
mza
96
6.5k
Site-Speed That Sticks
csswizardry
10
760
Designing for Performance
lara
610
69k
How GitHub (no longer) Works
holman
314
140k
Embracing the Ebb and Flow
colly
86
4.8k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
367
26k
Facilitating Awesome Meetings
lara
54
6.5k
Transcript
None
@listochkin
<3 Ruby
No Ruby today
None
Rust Favorite language Ruby Extensions WASM
RustFest Workshops Intro Advanced Adoption
Mozilla Travis CI Samsung Chef CloudFlare Evo
Steve Klabnik lislis Ingvar Stepanyan Paul Colomiets
Rust is for Rubyists
http://bit.ly/rustfest2017ruby 50% off
I’ll talk about my work
None
Bot building system
None
None
Conceived about 6 years ago
We’ve been building bots before it was cool
Bot
IRC WoW bots Web Scrapers
Chatbots
Conversational Commerce
A bot pretends to be a human
Makes stuff happen
Emphasis on “Human” aspect
Wit.ai API.ai Motion.ai Microsoft Bot Framework Init.ai <Whatever>.ai
Can you <deliver> <my order> to <my work address> instead?
What to do when there’s no match?
Guess Bail out
AI + Human
Facebook M 70% required human
Leaky abstraction
Bots are Dead End of story
“Human”-like is a red herring
How vs What
Goal Scenario Guidance
Humanize later
Scenario
Workflow
Do X Do Y Do Z
Customer case: PeopleReady
Workforce matching
“Uber for Workers”
I’m a construction worker I’m free next Friday
I need 5 extra construction workers on site next Friday
Phone Decision Tree
“If you need X press 1”
IVR “Interactive Voice Response”
Step 1: Static Tree
Step 2-∞: Refinements Shortcuts A/B tests
Identify a person/role Reduce tree
Known pending request: top-level notification
Global shortcuts
Keywords, numbers, etc.
Bot-to-Bot IVR-to-IVR
Pass context
Session storage HTTP/SIP headers Forward DTMFs
Dumb Boring
Efficient Predictable Less frustrating
None
Workflow Builder Execution Services Operator Apps Reporting
Many competitors Not a winner-takes-all
Lessons Learned
1. Providers
Slack, Messenger, Telegram, Skype, Email, sms, HipChat, WeChat, LINE, etc
The more the better?
SMS Phone Email
No middle man apps Brand <=> Phone
2. Execution model
State Machine
State Machine as Data
Flow Tree => Data Structure Executor
State Machine as Code
Flow Tree => Data => Code
Separate Infrastructure Separate Deploys Separate Load
AWS Lambda
3. Steps are the key
if … else Choice 10-15
Smart Steps 100s
Phone Transfer
Transfer to X Backup number? Voicemail? Transcode?
Feature discovery
Higher quality Bots by default
Programmable steps
Flow => Step
Step = Logic + UI both programmable
Custom JS Step
4. Service Flows
Pick up the phone Identify a User Voice Choice …
Number Provisioning API Adapter Reporting Events Hide Secrets …
Service flows Global Events with ACL
5. Dogfooding
Everything is a bot
Admin tools Monitoring Reporting Billing
6. UI and Reporting Sell Platform retains
UI! So many capabilities!
“Can you set it up for us?”
Typical Customer: sets thing up once receives reports
7. Reporting!
SQL Server OLAP Direct access
Let customers play with the data
8. AWS Lambda
async (req, context) => { ... }
Node 4.3 (6!) Python Java
Use Large
1 request = 1 new process
Dynamo & Redis :) Mongo, SQL :(
Dynamo is amazing!
9. Voice
Voice is inherently hard!
Twilio Best API and DevTools Still hard :(
Larger customers need cheaper voice
TwiML Twilio SIP + Custom Voice Direct SIP-to-SIP
SIP + RTP
SIP Commands & Events Bi-directional
RTP - audio
Often Arcane practices
Huge Risk
Opens doors
10. Small things Big Impact
Haiti Earthquake SMS Fundraiser
Suicide Prevention Hotline
Washington DC city bot
Roadside assistance Match-making Marketing Customer support Sales
OneReach
29-30 RustFest http://bit.ly/rustfest2017ruby 50% off
29-30 RustFest http://bit.ly/rustfest2017ruby 50% off 2 free tickets today