AI Hub: Anonymize. Redact. Escalate.
- Ana Inés Urrutia
- 20 abr
- 5 Min. de lectura
Once in a while I had a walk through AI hub in Power Platform and see what is new, or what calls my attention that maybe before wasn't s attractive.
I had a look and found something very sexy: a new prompt. For those who are not aware of AI prompts in Power Platform, these are nice ways to use AI for tasks that may require some additional work, but BEWARE, with great power comes great responsibility, so remember to use the Responsible AI practices (they are linked at the end of the article in case you want to know more).
Today I was diving into the Prompts section, which is always interesting and I found a hidden treasure. Let me guide you...
Ok, so everything starts in the AI Hub and then selecting prompts
After AI Hub opens, you can see some options. I was curious so I had a look at all the templates Microsoft provides, recommend 100000%, not only to use them in your apps but also to improve your prompting skills it's definitely helpful.
Below you can see a range of options, and todays pick is Text anonymization and redaction.
The idea of this prompt is to take personal identifiable and sensitive data from a person/business/project/you name it, anonymize it and redact a new text following those rules. Ensuring that this practice takes part of your processes guarantees certain protection for the subject. Let's put it simple - and always HR of course - if a candidate applies and we want Hiring Managers to evaluate the match between job and person the only thing the Hiring Manager needs is the data: years of experience, education, previous roles and eligibility to work in the country they are applying. It's clear that someone should be able to identify the candidate, most likely a person that is in a support role within the process - e.g. recruiter.
While the prebuild prompt from AI Hub provides a suggestion of the prompt instructions, I wasn't really happy with it. It was too plain and didn't add a lot of value to me. Maybe I'm too picky - but better to improve than complain. This is the defaulted/prebuilt prompt:
Given a that contains sensitive information such as names, locations, or personal details, develop a process to anonymize or redact this information. Ensure that the privacy of the individuals is protected while preserving the overall meaning and coherence of the text.
So I tried to level up and improve it - I did have some help from an AIssistant to draft the prompt. See below the new version:
You are a summarization assistant for recruiters.Given the following candidate application , generate a short, anonymized summary that highlights the candidate's relevant skills, experience, education, certifications, and accomplishments.Do not include any personal identifiers such as name, gender, age, date of birth, nationality, marital status, or photos. Avoid any phrasing that could reveal these details implicitly (e.g. “she has 10 years of experience” → “the candidate has 10 years of experience”).Focus on:Years of relevant experience
Key skills and technologies used
Achievements or results (with quantifiable outcomes if possible)
Education and certifications
Languages spoken (without specifying origin)
Work eligibility or relocation info (in general terms only, e.g. “eligible to work in the EU”)
Output a concise, professional summary in neutral language suitable for recruiters to review candidate qualifications without bias.
To make sure that the prompt works accordingly, I recommend running a couple of test with random data as you can see below.
What I want to do with the prompt is to add a field in an existing model-driven app where I have all the candidates record. Big part of the recruitment process relies on Hiring Managers so what we need is to make sure we remove personal data as much as possible. The challenge is that being aware or not, we evaluate and categorize things based in our own experiences - this is what is called the unconscious bias - and I'm positive you do it every day.
I have an app that is called AI Summarization - Denisa is the mastermind behind it - and it already retrieves a lot of data from the candidate profile - more info here: AI Models + AI prompts: CV screening with Power Apps and Power Automate. We will add a field in the candidate form that display the result of the anonymization process.
This is the overview of the app I just mentioned:
In the candidate form we have a new field that displays the AI summarization, let's see what the summary looks like:
And lastly - have a look at the beauty of AI prompts and the combination between thinks that work well together.
With this anonymized summary what we will do is to send it to the Hiring Manager so they can assess the match between person and job opening without having the bias I've been talking about.
I created a simple Power Automate flow, see below. First trigger is When a row is added, modified or deleted. In out case we don't care about rows being deleted or added (maybe added yes, but let's put that on hold) so we proceed with the change type as Modified.
In the second step I added the Send email action, which includes the Anonymized summary - which is what the Hiring Manager will receive and make a decision moving forward.
This flow will send an email to the Hiring Manager every time the anonymize and redact feature is used in the recruitment model-driven app. Managers receive an email like the one below:
The email summarizes the profile + calls to action - in this case the actions are performed manually and are not part of the scope of this article but if you are interested we can see how to automate that part as well :)
What are your thoughts on the anonymization prompt? Have you tried it yet?
Comments