Make + ChatGPT to receive and evaluate Upwork jobs

Prologue

I recently decided to revamp my Upwork profile and start looking for clients on the platform. However, searching for relevant jobs on Upwork can be incredibly time-consuming—it's almost like having a separate part-time job! 
With Upwork, it’s just not really cool to use just filters, because:1) A lot of budgets defined by clients are still placeholders2) Clients often publish projects in the wrong category
When I fetch projects from Upwork in Sales&Marketing category to my inbox, the biggest part of them are still not relevant. To evaluate every project, I needed to read the description. So I decided to hire ChatGPT to make the project evaluation from 0 to 10 according to the description. So I’d receive in the inbox all projects, which are more than 7, let’s say. 
To set up this automation, I used Make (formerly Integromat).  

My scenario in Make

RSS module setup

To begin with, I created an RSS module. In the RSS module, I simply entered the link to the Upwork job feed, where I prior set up some filters:

Upwork filters

OpenAI module setup

The RSS module takes each job from the feed and sends it to OpenAI for review. For the OpenAI module, I used the following settings:Select Method: Create a prompt completionModel: text-davinci-002Prompt (be as detailed as possible):
`Here's the project name on Upwork: {{1.title}}Here's the description: {{1.description}}
Please rate how this project is relevant to my skills from 0 to 10, where 0 - not relevant at all, 10- super best fit.
Who am I: I am an email marketing and cold outreach expert. I am based in Poland. I work with clients from Europe and USA.I am interested in jobs where client needs to set up cold outreach system, or build email marketing system. And I am also interested in marketing automation, where I need to connect CRM to some program via Make or Zapier.I work with ActiveCampaign, Lemlist, Instantly, and most email tools. I don't work with Klaviyo. What I can do:- setup a domain for cold outreach - warm up the domain and mailboxes- integrate the CRM system so that responses/bounces/unsubs are tracked via CRM- create hypotheses and prioritize portraits/personascreate emails and sequences- launch email campaign- track results and provide the feedback- monthly testing domain reputation- provide automated and regular email marketing campaigns- use workflows for segmentation and best engagement- track open rates/reply rates/so on for the best deliverability
Just return me a pure number of how this project is relevant without any other words.`

Email module setup

The third module in my integration flow is email. After OpenAI reviews and assesses the job, I receive an email with an assessment mark and the job title in the subject line, along with the job description in the content.

Here are the settings in my email module.

And here is how an email looks:

Email I get from ChatGPT

As you can see, the process worked. However, is the quality of search and assessment acceptable, and do I consistently get only relevant projects?

Results review

Well, I got almost 250 emails (meaning 250 jobs) in 8 days.
10-rated jobs by Open AI. There are only 9 of them. My review:● I’d give a 10-mark to 3 jobs; they are super-relevant and fit my request.● Half of the jobs were completely irrelevant: required php knowledge, called calling experience, experience with particular tools that I hadn’t mentioned in my request, needed someone Philippines-based (though I mentioned to OpenAI that I’m based in Poland)● A few jobs were partially relevant. However, they weren’t the top priority for me at all. So I should teach Open AI not only the task I want to do but also the requests I don’t want to take.
9-rated jobs by Open AI (43 items) My verdict:● Only 4 jobs match their score and are highly relevant.● Again, many jobs require skills that I haven’t mentioned (Amazon, publishment, graphic design, copywriting)● Many small jobs (fixing a tiny issue) or low-paid ones. So I decided to set up a minimum rate/fixed price.● Lots of skills/conditions that I should include in the not-to-consider list for the future. 
To sum up, 9-mark jobs were less precise and relevant than 10-rated ones. I thought this tendency would continue, and 8-mark jobs would be even less applicable. However, the results surprised me.
8-rated jobs by Open AI (159 items) My verdict:● Quite a few relevant jobs that I’d give 8-10 to.● Of course, many irrelevant descriptions.● However, I noticed that OpenAI tends to give 8 rather than 9 or 10 to appropriate jobs. And relevant jobs with 8 marks aren’t less relevant than fitting ones with 9 or 10 rates. Feels like it marks jobs 8-10 randomly.
Now let’s see what kinds of jobs were rated 0 by OpenAI.They are truly irrelevant. For example, many of the 0-rated jobs required Klaviyo knowledge (I mentioned in the prompt that I don’t work with it). To be fair, I must mention that some jobs with Klaviyo were also rated 8-10.
Good work done here!

Key takeaways:
1. You can set up an automated scenario that will make ChatGPT rate Upwork jobs.2. OpenAI does a great job in cutting off irrelevant offers, and it relatively well finds 10-mark jobs. 3. On the other hand, OpenAI has hard time rating jobs 1-7 (there were few such results). So it identifies extremes well and hates relativity.4. For future experiments, I’ll ask OpenAI to rate the projects as `yes` or `no` and show only yes-ones to me.5. Also, along with must-have requirements, it’s important to have must-not-have conditions (the more - the better).6. ChatGPT needs feedback on how I would rate this project.7. For some projects, it was still purely random.