How I analyzed 3,000 HRIS platforms in 48 hours with o1-preview and Perplexity and Cursor

13 min readSep 22, 2024

Last Saturday (Sept 21st 2024), I set myself a challenge: to identify as many HRIS (Human Resources Information System) platforms using generative AI as possible, all over the world.

But how do you do that? There are thousands of startups and companies in the HRTech sector. Using generative AI, of course! I used a mix between Perplexity and o1 to achieve my goal. As we’ll see later, it’s not perfect. But it’s the best method I’ve found for automated intelligence at the moment!

I’d like to point out that, although I’m a computer geek, my knowledge of data science is more than limited, which makes the exercise all the more impressive.

The results?

Over 3,000 HR companies analyzed
1,400 identified as HRIS
709 using AI
166 specifically using generative AI!
An application to visualize the HRIS analyzed: see it for free here
3 python scripts
3316 lines of code

And all in less than 25 hours (calculation at the end of the article) and all by myself! Finally, with my new AI companions, Perplexity and o1-preview from OpenAI.

Here’s the final software I created, available free of charge here.

How do I calculate the estimated time?

Throughout the article, I give an estimate of the number of hours it would have taken to achieve the result without the help of generative AI at each stage.

To calculate the number of hours it would have taken, I’ve based myself on an average of 50 lines of code per 7-hour day, which I think would be normal for an average IT developer.

What is Perplexity?

Perplexity is an artificial intelligence-based search engine, similar to ChatGPT, but far more powerful when it comes to precise, detailed searches. What sets it apart is its ability to crawl the web in real time to provide comprehensive answers, often with reliable sources to verify directly. At the moment, Perplexity uses Meta’s Llama 3.1, more specifically llama-3.1-sonar-huge-128k-online, a state-of-the-art model designed to handle huge volumes of online data hyper-efficiently.

In addition to being a search tool, Perplexity offers an API (programming interface) that enables other applications to easily integrate with its technology to automate information gathering. This allows companies or developers to leverage its search capabilities to get real-time answers directly into their systems or applications.

You can try this application here, free of charge.

What is o1 Preview?

o1-preview is the new artificial intelligence model launched by OpenAI. It is designed to excel at complex reasoning, meaning that it can solve difficult problems by taking the time to think through each step, the so-called “Chain of Thoughts”, much as a human would; indeed, it is on typical examples of human thinking (in text) that this model has been trained. This makes it particularly effective for tasks such as mathematics, science, and especially computer coding, where it outperforms all other models (and many humans).

For example, it has shown excellent performance in online programming competitions, ranking among the top participants. This model is also capable of generating and optimizing code, resolving bugs and assisting developers with their projects.

I can immediately hear people crying anthropomorphism, so yes, these LLMs don’t really sound like humans, these systems are only artificial, only simulating reasoning. I think this is a pointless debate. I’d still like to point out that these models were specifically designed to reproduce the way a human thinks, and it works. So in the end, whether it’s “true” human reasoning or “false” reasoning simulated by a machine, we’ll take the best result, right?

Parenthesis aside, it turns out that Tomorrow Theory’s OpenAI account dates back to 2020, (and yes, I was among the early adopters!) and offers us certain advantages, including early and unlimited access to new models, here from o1-preview.

Workflow

First things first. I’m doing a lot of research into HRIS right now, and I’m wondering how generative AI is going to radically transform this type of software, from recruitment and payroll to learning and talent management.

So I had an idea: why not use Perplexity to automate the search for start-ups and companies innovating with generative AI?

1- Corporate harvest

Time spent with AI: 3h

Estimated time without AI: 27h — 2h for manual information gathering, 25h to extract 3000 companies from unstructured data, 30seconds per company

First step: Gather as many names of HR organizations as possible. I started with the French ones, looking for all the exhibitors at HRTech Paris 2024, then the companies making up LabRH, then the European HRTech, the US, Asia, Oceania, Thailand, Singapore, Russia and even China! Then I went to startup incubator sites like YCombinator, HRIS listing sites, etc…

Here are the main sources I used:

This step was the most tedious, as autonomous AI agents don’t yet exist, so I had to do this sourcing partly by hand.

First I went to one of the above sites, copied all the pages mentioning companies without sorting, then I gave all this unstructured data to o1 and it did a surgical extraction of company names, up to 200 company names extracted without error in one go, and I didn’t try any more!

https://www.mapofhrtech.com — Carte du HRTech Chinois

The most complex to extract was the Chinese HRTech. I had to use this infographic, extract the text with the image to text Brandfolder application, and give the raw text (in Chinese) to o1. He managed to come up with around 180 names!

Once that was done, I found myself with an excel file containing over 3,000 lines and 3,000 company names.

But there’s a problem: I’ve got quite a few duplicates!

2- Dedoubling

Time with AI: 2min

Estimated time without AI: 3.5h (25 lines of code)

So here I am with my 3000-line file, and quite a few duplicates. I want to write a script to remove these duplicates, and to do this I need to pay attention to case, among other things:
For example, Workday and workday must be considered as the same company, despite the capitalization of one and not the other.
So I decide to write a script, or rather, I ask o1 to do it. And boom, in 30 seconds, I’ve got a 30-line script that works on the first try and even sorts all these little people in alphabetical order!

Prompt o1:
I have 3000 company names in an excel file (only one column: A) I have a lot of duplicates, I would like to clean up this file. There’s some uppercase/lowercase stuff, etc… make sure you remove all the duplicates. Be careful, there are companies in Chinese and Cyrillic characters. Write a python script for this.

Well, not bad, other AIs like Claude 3.5 already allow you to generate simple scripts. But hey, I’m happy, my new file now contains 2959 companies instead of 3024, that’s 65 duplicates.

Ok, now that my source file is clean, it’s time to get down to business: automated monitoring.

3- Automated monitoring

Time with AI: 14h — 1h to write the script, 10h to wait for the script to finish running.

Estimated time without AI: 560h — 500h: at the rate of 5 min per company, not counting the days spent thinking and 60h coding the script by hand (432 lines of code)!

Here’s the interesting part! The part where Perplexity comes into play. Here’s the plan: for each of the company names, I’m going to ask Perplexity to give me as much info on the company as possible to determine :

If it’s an HRIS (in the 3000s, a lot of HR consulting firms, without HRIS software, and I’m not interested in that today!).

If yes: we enter the company in the list and ask Perplexity to :

Collect company URL
Summarize the company and its HRIS solution
Collect data (country, founding date, HR sector, etc.)
Determine whether the company uses AI
Determine whether the company uses generative AI specifically
Collect sources proving this

If no: we don’t register the company and move on to the next one.
Once again, I ask o1 to write the script for me, and it takes me 1 hour of iterations to get what I want. In the end, I end up with a 432-line script, which gives me an Excel file and a JSON file with 1,422 companies offering HRIS software!

In this script, I have 2 prompts for Peplexity:

Prompt 1: which determines whether the company has an HRIS and collects the company’s info
Prompt 2: if it’s an HRIS, then determines whether the HRIS uses AI or generative AI

Prompt 1 Perplexity :

# The first prompt template with founding date
 first_prompt = f'''
Provide the following information in JSON format, JSON only, without explanation:
1- Does the company offer a software or technical solution for HR?
Yes
No
2- HRIS categories
Choose from the following categories (multiple choices possible if applicable):
- Recruitment and selection of candidates
- Onboarding and integration
- Training and development
- Performance management
- Personnel administration
- Employee engagement and well-being
3- Company size
Select one of the following standardized sizes:
- Startup (less than 50 employees)
- SME (Small or Medium Enterprise, 50 to 250 employees)
- ETI (Entreprise de Taille Intermédiaire, 250 to 5000 employees)
- Large company (more than 5000 employees)
- Multinational
4- Country of origin
Use the official name of the country in French, for example :
- France
- United States
- United Kingdom
- Germany
- Spain
- Canada
- Australia
- etc.
5- Company creation date
Format: Year (e.g. “2010”)
6- Summary of HRIS software and added value
A concise text describing :
- The company's main activities
- Key products or services
- What sets it apart in the market (innovation, sector specialization, use of AI, etc.).
7- Company website URL
Full format, e.g. :
https://www.nomdelentreprise.com
This is the company: “{company_escaped}”.
Note that there may be several companies with this name related to Human Resources Information Systems (HRIS) or Human Resources (HR). Please perform a thorough search using the terms “HRIS” and “HR” to identify the relevant company.
It is very important to provide the company's website for verification. Or failing that, a reliable source confirming the company's existence. 
Please note that the “website_url” field must only contain a valid link starting with “http://” or “https://”, nothing else, no explanation if it's not the official link.
=======
JSON format :
[
  {{
    “company_name": ‘’,
    “is_hris": ‘’,
    “categories": [
      “”,
      “”
    ],
    “company_size": ‘’,
    “country_of_origin": ‘’,
    “founding_date": ‘’,
    “summary": ‘’,
    “website_url": ””
  }}
]
===========
Important reminder : The output format must be JSON, **only JSON**, without any explanations, comments, remarks, notes, or additional text. Do not include **any** text before or after the JSON. Your response must begin immediately with the '[' character corresponding to the start of the JSON array.'''

Prompt 2 :

second_prompt = f'''{first_response_json_str}
Does this company use AI and more specifically generative AI?
=======
JSON format:
[
  {{
    “use_ai": ‘’,
    “explanation": ‘’,
    “use_generative_ai": ‘’,
    “source_urls“: [””, ‘’]
  }}
]
1- URL sources should be links to articles, website pages or documents that confirm the company's use of AI.
2- Only include sources if the company uses AI.
3- If the company uses generative AI (this use must be explicit), then use_generative_ai = “yes”, otherwise “no”.
4- If the company does not use AI, then use_ai = “no” and the other fields remain empty.
Important reminder : The output format must be JSON, **only JSON**, without any explanations, comments, remarks, notes, or additional text.
Do not include **any** text before or after the JSON. 
Your response must begin immediately with the '[' character corresponding to the start of the JSON array.'''

EXCEL File:

JSON File:

[
    {
        "company_name": "Fusion5",
        "is_hris": "Oui",
        "categories": [
            "Administration du personnel",
            "Gestion des performances"
        ],
        "company_size": "ETI (Entreprise de Taille Intermédiaire, 250 à 5000 employés)",
        "country_of_origin": "Nouvelle-Zélande",
        "founding_date": "Inconnue",
        "summary": "Fusion5 propose des solutions de gestion des ressources humaines (RH) et de la paie pour les entreprises, couvrant les processus de gestion du personnel et des performances.",
        "website_url": "https://www.fusion5.com/",
        "use_ai": "oui",
        "explanation": "Fusion5 utilise l'IA pour des solutions de gestion des ressources humaines et de la paie, ainsi que pour des services de données et d'IA.",
        "use_generative_ai": "oui",
        "source_urls": [
            "https://www.fusion5.com.au/microsoft/microsoft-copilot/",
            "https://www.fusion5.co.nz/data-and-ai/"
        ]
    },
    {
        "company_name": "2Interact",
        "is_hris": "Oui",
        "categories": [
            "Recrutement et sélection des candidats",
            "Onboarding et intégration",
            "Formation et développement",
            "Gestion des performances",
            "Administration du personnel",
            "Engagement et bien-être des employés"
        ],
        "company_size": "Inconnu",
        "country_of_origin": "Non disponible",
        "founding_date": "Inconnue",
        "summary": "Non disponible",
        "website_url": "Non disponible",
        "use_ai": "oui",
        "explanation": "",
        "use_generative_ai": "oui",
        "source_urls": [
            "https://www.linkedin.com/company/interactsoftware"
        ]
    }
]

Now that’s crazy! I’m still amazed at how far I’ve come in such a short space of time!

Here I’ve identified 1,422 HRIS, including :

709 using AI
166 specifically using generative AI!

Well, I consider the reliability of the results at 90%, which is good but not great. Still, it’s nice to know that I’ve spent 546 hours less on this problem than it would have taken me a few months ago!

I still saved 97.5% of my time!

But then again, 90% reliability isn’t quite there yet, even if a few HRIS startups in generative AI must have slipped my mind, I can dig out the ones I’ve already got.

4- Targeted improvement of results

Time with AI — 1h30: 30 min to write the script, 1h to wait for the script to finish running.

Estimated time without AI — 60 hours: 20 min per company, and 5h to code the script by hand.

Here I want to take the 166 HRIS using generative AI and dig a little deeper. To do this, I need a new script that will do several things:

Resume the 166 startups one by one
Fetch and copy information from the collected sources talking about the company’s use of generative AI
Use Perplexity to specifically search for how the use of generative AI applies to this company.
Reconcile the data from the sources and Perplexity using the power of o1 and provide a more detailed report.

I’m using o1 again to generate the script, this time 250 lines long.

Here’s the result of an advanced search with Perplexity and o1 for the open-source Odoo software, for example:

L’entreprise Odoo utilise concrètement l’intelligence artificielle (IA) générative dans les Systèmes d’Information Ressources Humaines (SIRH) pour améliorer l’efficacité et l’automatisation des processus métiers. Voici des détails sur la manière dont Odoo intègre l’IA dans ses solutions SIRH :

1. Process automation: Odoo automates many repetitive tasks and processes, helping to save time and reduce errors. Workflows can be customized to schedule automatic activities, generate real-time reports and use artificial intelligence to optimize business operations.
2. ChatGPT integration: Odoo has integrated ChatGPT, an advanced language model, to offer real-time assistance with fast, accurate answers to user questions. This integration enables companies to benefit from intelligent support for their business operations.
3. Human Resources Management: Odoo offers advanced functionalities for human resources management, including recruitment, training, talent management and employee engagement. Odoo’s integrated technology solutions, such as e-recruitment platforms and applicant management systems (ATS), help to improve efficiency and talent management.
4. Flexibility and customization: Odoo is a customizable open source application suite that enables companies to meet their specific needs. Odoo modules and applications can be extended through additional downloads, offering great flexibility for businesses.
Software used
- Odoo : Suite of open source applications for business management.
- ChatGPT: Odoo’s advanced language model for real-time support.
Sources
1. Ringover: https://www.ringover.fr/blog/odoo
2. Datascientest: https://datascientest.com/odoo-tout-savoir
3. Weopensoft: https://weopensoft.com/odoo-chatgpt
4. Tech For Retail: https://www.techforretail.com/solutions-et-produits/ressources-humaines-recrutement/
5. Odoo-BS: https://www.odoo-bs.com/fr_FR/odoo
These sources provide detailed information on how Odoo uses generative AI in HRIS, as well as on the software and modules used to improve the efficiency and automation of business processes.

So now I have a real report for each of the 166 companies!

I could stop there, but I want a way to share this knowledge. So I decided to make a little application to visualize all this data.

5- Data visualization web application

Time spent with AI: 3h

Estimated time without AI: 110h for 781 lines of code

Once again, I’m using o1 to help me in my task. He’ll be in charge of architecting my solution, but I’ll choose the technology:

It’ll be NextJS, a techno mixing React.js and Node.js for those in the know, a MongoDB database and deployment on Vercel.

Once I’ve finished my prompt, I ask o1 to generate the application code for me.

I use Cursor, an AI code editor, a fork of VSCode released this year that I already can’t live without, it integrates Claude3.5 and lets you code at lightning speed.

In just 3 hours, I’ve been blown away, and I’ve come out with a perfectly functional program with sorting, filters and all the data from my automated watch!

You can test it here for free: https://sirh.tomorrowtheory.com/

6- Rédaction de cet article Medium

Temps estimé sans IA : 4h pour 303 lignes de texte !

Ce post a été entièrement rédigé sans l’aide de l’IA ! Il faut bien garder un peu le plaisir d’écrire ;)

Estimated time without AI

You’ve all been waiting for this: here’s the table summarizing the time saved with o1, Perplexity, Claude 3.5 and Cursor.

Are you ready? Ready to see your productivity, your business and the whole of society transformed by generative AI?

Follow us at Tomorrow Theory, as we anticipate these major changes, keeping in mind that humans must remain masters and actors in their own lives, with awareness and responsibility!