Full transcript
0:00today I'm going to go over a referral
0:02agent now the referral agent I've built
0:05basically takes any new customers that
0:08we've had with my company triggery and
0:11reaches out to that individual's team
0:14and says hey your team has signed up why
0:16didn't you also come and sign up as well
0:19the idea being is it takes one user and
0:21turns it into three if we can do so so
0:25let's have a look so here we are we're
0:26in um na10 this is just a like I said
0:29it's like a zapia and make on steroids
0:32because we can add some codes we have
0:33some great agent capabilities in here
0:36and so on we're using a web hook now and
0:39so as you can see here I'm basically
0:40taking a web Hook from our
0:42authentication provider which is clerk
0:44authentication just means it manages
0:46people signing up to our website and
0:48what basically happens is as a new user
0:50signs up we receive that data via a web
0:53Hook from there we start to then
0:55basically clean it so we get their first
0:57name last name and email address from
0:59the Web book so clean the data and then
1:01we send it through to this um AI agent
1:05inside of this AI agent we've given it
1:07the following prompt go find me one to
1:09two people from this company we give it
1:12the email address don't find me this
1:14person so that person being the
1:16individual that's already signed up with
1:18my company we've also given it what we
1:20call like a system message and this
1:22system message says you're helpful uh
1:24assistant that will enable me to do some
1:26prospecting and call on the People
1:28Search tool I'm only interested finding
1:30people from directly leveling up within
1:32the following parameters and then I give
1:33the following titles here we're using I
1:35think we're using just yeah we're using
1:3740 as the model we don't need to use I I
1:39do find that Claude these days is a
1:40little bit better but obviously GPT is a
1:42little bit cheaper and if you're doing
1:44it for more just kind of mundane tasks I
1:46tend to use GPT for that we're using
1:48Apollo here and we're using their people
1:50search API now the reason I haven't just
1:52done this via code is we can also start
1:55to upload what we call like Memories
1:57right to the agent and here what I tend
2:00to do is I start to upload memories I'm
2:02using like a vector database or I
2:04sometimes use like a postgress as well
2:06and I'm giving it information on who who
2:09it should look for right and so then
2:11what I mean by that is if there's like
2:12five or 10 potential people to go after
2:15it looks at our current database of
2:17people that are clients and works out
2:18the best one that would be appropriate
2:20for them and you can just think of with
2:22the prompt to get it there then what
2:23we're doing is with saying Hey I want
2:24you to return this data in a structured
2:27Json format and so we've got full name
2:30LinkedIn URL job title email and then
2:33the domain so what then happens it gives
2:36us that Json and then we basically say
2:38take that Json and we split it out into
2:40multiple different uh fields from there
2:43we then start to send this data through
2:46we're using popera in this example now
2:48to get their email address once we have
2:50their email address it then sends that
2:52data through to Smart lead now one of
2:55the key things that I'm doing here is
2:57Prospero can't handle like five people
2:59at one time so we do this thing here
3:02called Loop over items which is
3:03basically splitting into batches the
3:05batch sizes of one so what that
3:07basically means is we have five people
3:08coming through here it stops them here
3:10and only sends one through and that as
3:13that one through happens it then sends
3:14the next until it's done or five where
3:16then sends it up to here where we're
3:18also doing another loop because smartly
3:20can't handle large quantities at one
3:22time and we send it one one at a time
3:25through into smart lead now this is kind
3:27of more I would say like an AI automate
3:30a little bit of like agent kind of ask
3:32if you're really doing the memory side
3:34of things then yes but it's an
3:35incredible workflow and automation build
3:38that's super super simple I'm going to
3:39leave the Json of the file Below in the
3:42description of this video so you can
3:43just upload this simply by coming here
3:46right I think um if you create new
3:49projects workflow I think you can then
3:52import from file and then away you go
3:54you can upload everything that you're
3:55wanting to do from there but as you can
3:56see I've got a lot of different agents
3:58that I use I continuously adding agents
4:01here because I'm a big believer in
4:03taking my business trigger fi to you
4:06know 10 20 mil with as few people as
4:09possible and I feel like agents of the
4:10way to do that I started the journey
4:13learning how to do low code and I still
4:14use na10 a lot and I'm going to spend a
4:16lot of time with na10 but I've more
4:17recently also started to do a lot of
4:19actual code development using things
4:20like cursor using things like wind surf
4:23us things like bolt vzer these these
4:25types of tools and I'm going to do some
4:27videos and tutorials quite long ones
4:30basically showing how I just build full
4:32micro SAS and SAS applications which
4:34include kind of agents in that as well
4:36so if you like what you're seeing like
4:38comment subscribe and follow on on the
4:40journey