Taken at Babycastles in NYC 10/11 NYC Botmakers Meetup for more info: https://www.meetup.com/botmakers/ I make no claims as to the completeness of these notes
3 days of full development (idk what a bot is to v0.1)
What is the weather in nyc? turn that into intent which the server can understand API converts chat into action and parameter which is sent back to sever and service the original request
collects information via FB messager which dumps candidate info into a slack channel which recruiters are watching
pre-built dialog agents
create entities (wired, conde nast, etc) and train new intents in dialog flow
create entities for all brands, train new intents to all the questions
how to: use slot filling (it’s like a step-by-step validation form) with user sessions to prevent cross-talk client-side validated field slot filling: User bio, user email
how to: persist send to a database or slack. Send profile info (from launch vehicle) and user submitted info (from dialog flow) to slack
generative text by collective probability map for different keywords
code editor and webhost live updating, forking, remixing allowed
takes HTTP post/get requests and output tweets using twitter API mad-libs kind of situation
bot.js starts server witter.js posts to twitter tracery.js builds tweet also, tracery config
github.com/dariusk/corpora lots of random shit enumerated into a giant dataset
took the tracery code out, put the corpora code in daily_tarot_bot
Free hosting! But your instance sleeps if it hasn’t been active in 5 minutes So you can’t use a cron
uptimerobot.com
you can highlight lines to ask what things mean
don’t post too often you’re responsible for the content your bot posts don’t @ people (unless they follow the bot)
Dev at clarif.ai @maxcell
CNN - convolutional layer model based on convolving image subsets and outputs confidence scores
pre-trained search build your own model
glitch bot for slack image lookup
@lanewinfield
where is the sound bite from? on fiver, lmao
pytube - import youtube but pytube can’t search, so use beautifulsoup4 moviepy
ffmpeg-> numpy,scipy, PiL -> ffmpeg random yt video + the meme -> automatically edited, moviepy -> twitter video -> frame commercial -> audio clips (song, scratch, voice-over) @robotpornaddict azure computer vision google tts + moviepy
https://github.com/lanewinfield/freezeframebot
started as a daily email & sms service then apps: alarm clock, weather + commute proprietary tech: poncho brain mostly editors, not developers preview window to mock up content before it goes live post conversations to slack to observe
diagram user paths through poncho’s content trees helps to understand how users interact with the bot
broadcast or send content as a follow-on to daily weather forecasts
define character bible
60% 7-day user retention 8 months after launch
personal notification secret sauce: emotionl connection designing through a characer visual design heavily used to support poncho’s character and content strategy
inauguration therapy engagement 100% increases in message exchanges be real and respectful leverage platform features to improve your ability to reach audience
be corrective be predictable be fresh
conversational interfaces can come from many forms you don’t need to start with ML/Ai/NLP don’t be afraid to experiment
find a good balance of visual and text
stay at the top of the feed
rundexter.com botmakers.org howdy’s botkit rivescript
workplace voice devices virtual influencer poncho everywhere (comics, horoscopes, partnerships) @ponchoIRL