Building an AI Email Escalation Agent 🤖 (alternative)

Sorting, classifying and even potentially replying to those.

Well, this video is for you.

This is a real problem.

I'm going to show you with a real solution.

You're going to build your first AI agent.

Check it out.

This is a real recent posting on Upwork.

and it can be great opportunity for you to see a real AI agent.

Example.

it's very beginner friendly.

The lead or potential client wants an automated email escalation intake mechanism.

Let me show you how to do it.

So I pretty much reduced this request to one job, this AI agent I'm building will have a specific job which is, let's say an AI employee.

its main job is turn every issue email into a review ready case.

Gmail is the input.

Hermes, that's the AI agent I'm using my ChatGPT subscription.

extracts the request and we use the policy rules that of course would be crafted along with the client to make the decisions the recommendations.

And those decisions are fixed policy rules like deterministic ones, to call for attention.

We have Slack as the communication channel.

There will be a dashboard as well.

I'm gonna show you in a minute.

And of course the Gmail itself will with the drafts and all Gmail connection Can read only the matching last messages and create drafts.

This is of course custom made.

for this example you can.

Change this quite easily.

Hermes doing a very specific job which is turn the messages into a of fixed fields with issue reference, deadline, missing information.

Of course I don't have all the information from the client.

course in this example, right.

The model then extract those facts.

but code sets the priorities.

Deterministic code.

an operation is blocked, it becomes priority one and it stops for human, review For this test, I have prepared six different made up situations, as you can see here.

From Slack.

I asked the agents to process it and report back to me with the urgent ones or not.

And we can also schedule this so the user or the client won't have to come here and ask every time.

It will still work on demand, but we want some scheduling, right?

And as you can see here, we got a response back on Slack, saying for instance that we have a highest priority P1 can show and see that in the dashboard as well.

The dashboard preserves the original evidence, as you can see, The policy rule and the outcome and it even drafts a reply back from your Gmail and it does that for all the other cases as well.

They're all classified P1, P2, P3 etc.

since we connected our Gmail, We even have the draft here.

It's not sent of course by design it's deterministic.

It cannot do it only if we click on send.

that is also custom made for sure.

So we solved this with an AI agent for you are learning your example using Hermes, GPT as the model with the subscription, Slack as the channel and Gmail connector.

You could also solve it for yourself using your own Claude Cowork or GPT work and this would become a skill running on a schedule.

Of course a client version would use their inbox rules database and security requirements.

But I intend to show here more real cases.

As you could see, this was a real Upwork job.

post and it's as real as it gets.

Comment what you think and thank you for watching.