Exclusive Interview with Henrique Milli, Creator of MidasCV: AI at the Service of your Curriculum | Generative ai certification google | Non-generative ai examples | Free ai courses | Turtles AI

Today on the Turtle’s AI Interview Space we have the pleasure of hosting Henrique Milli, creator of MidasCV. Henrique will explain how and why he developed this platform, designed to simplify CV creation and improve the user experience.

At Turtle’s AI we care deeply about the role of artificial intelligence in fostering and enhancing work, by streamlining personnel selection processes. In this perspective, MidasCV aims to highlight true merit and skills, contributing to a more effective matching between labor market demand and supply.

During the interview we will explore Henrique’s personal experience and his technological choices, seeking to understand how his solution can support a more transparent and efficient approach in the search for talent.

DukeRem: Hi Henrique, welcome to Turtle’s AI! Tell us about your educational background and the experiences that pushed you into the world of technology and innovation.

Henrique Milli: I would be just another John Doe if I weren’t originally from Brazil. My paternal side is Italian, but they had spent a couple of generations in South America until my father decided to search for a better life in the old continent. I arrived at the age of 10 in Manerbio, a small town in the province of Brescia.

I never really adapted to the climate and the people—who knows, perhaps if I had stayed in Brazil it wouldn’t have been any different. But being the weakest of the litter and the quintessential sacrificial victim throughout my adolescence filled me with a radioactive determination down to my bones.

At the age of 21, I decided to change my life and moved to Milan without a specific plan. I didn’t have much money in my pocket, but as any broke person would say: it’s all about how you use it. In fact, I managed to stretch my funds to rent a one-bedroom apartment and sublet the main room. With this trick—and not infrequently, a visit to “Daily Bread”—plus working as a Domino’s driver, I was able to support myself.

But it wasn’t enough to prove to everyone that they were wrong to exclude me instead of becoming my friends, and it wasn’t enough for me to believe I was worth something after all those years of hearing the contrary. That’s when I started working on Upwork for symbolic fees as a WordPress developer.

DR: Was there a specific moment or experience that inspired you to embark on this project? What motivated you to create MidasCV and how did the idea come about?

HM: Shortly after I started working on Upwork, a startup for which price was a priority decided to start collaborating with me almost full-time. Since then, I have always worked remotely, and it has been the best thing that ever happened to me. Moving from project to project, I not only became a better programmer but also a more convincing candidate.

In fact, when you start asking for market-average rates or a little more, it suddenly becomes much harder to secure a real interview, as recruiters often filter you out. Thus, I received many “nos” that helped me understand how the recruiting process works and how to present oneself as a strong candidate.

The problem is that you must dedicate many hours to a single application, where, even if you are the best, the chances of success are very low. For example, every position requires a specific skill set and if you create one master CV listing everything you can do, you risk the recruiter dismissing you because they don’t immediately find what they are looking for, or because when they do, among other things, they become convinced that you aren’t very good at any of them.

For this reason, each time you have to filter your experiences and projects to align with the position, creating a more digestible presentation for non-technical personnel. But, as mentioned, this takes a tremendous amount of time. So I developed a tool that could download my data from LinkedIn and, leveraging LLMs, populate a clear and attractive template for recruiters with concise and relevant information in just a few seconds.

When, seven years ago, I decided to change my life, I wanted to do it big—not just settle for being a “code worker”—so I asked myself: “What if I turned this tool into a product? Will others need it enough to pay for it?”

Thus, MidasCV was born.

DR: What skills have you developed throughout your career that you consider fundamental to bringing MidasCV to life?

HM: The most important of all is keeping a cool head. If you’re not funded, if you have no family money, and if you have no means to support yourself except by working, dedicating most of your time to developing and marketing a product is an extremely stressful activity. You risk your well-being and must make many sacrifices to scrape every last drop from your miraculous jar of juice. Creating a successful startup from scratch, without money, without an MBA from Bocconi, and without anyone believing in you is, in every sense, a miracle.

We human beings are capable of miracles, but throughout history very few have bothered to remind us of that. I believe that a frightened and insecure human being is, sadly, a more remunerative resource in traditional production structures. Believing with all my might that I can make it is by far the most important skill I cultivate every day.

Technical skills provided me with the indispensable foundation to face the challenge of launching a product, but if I hadn’t started believing in myself, I would never have acquired coding skills—and above all, I wouldn’t be persevering in building a company, marketing the product, finding funding, and doing my best every day despite results that are morally hard to digest.

DR: How does your vision for the future of recruiting and CV management reflect in the design of the platform?**

HM: I don’t see the future of recruiting much different from the present. Technologies may change, but the fundamentals will remain the same. The demand for white-collar jobs might drastically decrease and/or the number of recruiters needed to satisfy the market could diminish thanks to digitalization and AI.

Nevertheless, I expect that fundamentally the candidate will always have the responsibility to decide what and how to present themselves. Today, this responsibility directly translates into creating and sending a CV, and I do not foresee this fact changing significantly in the near future.

Even if we were to see a preference for proprietary forms and platforms, we could add features to MidasCV to populate those fields with the same knowledge and technologies we already use today to generate CVs.

Even if the recruiting process were to integrate directly with LinkedIn—which we currently use as the repository of all the candidate’s experience to generate tailored CVs—we could leverage existing infrastructures and know-how to turn Midas into an AI career advisor.

I can still add one word about the design: simplicity. Midas was born to solve a problem intrinsically linked to the information overload typical of our world. The interface, in fact, uses the minimum number of visual components, and the overall experience is designed to require minimal cognitive effort from both the candidate and the recruiter.

Not because one or the other has particular difficulties, but because we live in a world where cognitive overload is not yet a topic to which the public is particularly sensitive, though it is something we all certainly feel.

DR: Considering that MidasCV is still emerging, what are the development prospects and how do you envision its evolution in the market?

HM: Midas is the first product I am creating as a company and I’m not exactly sure how it will evolve. Every day I wake up, have breakfast, exercise, take a shower, tie my tie, and start testing ideas. These can work, fail, or not deliver the expected results.

As an emerging magazine, I’m sure you understand the mental flexibility required to see several possible futures along the spectrum, without yet knowing which one reality will collapse into—thus having to invest a little in all of them to understand which direction to take.

I confess that one of the ideas from the very beginning was to focus on the B2B world and provide this tool to those working in recruiting who face the problem of having to present their candidates in the best possible light to their end clients several times a day.

This is a use case we definitely want to propose to the market, because even though for a freelance developer it can be very useful and cost next to nothing compared to the enormous number of hours saved and the economic benefits obtained, everyone hates having to pull out paper for another subscription—and convincing people on a large scale of the benefits is very onerous.

But I still want to keep both paths open and evaluate where to invest based on the data.

DR: How did you organize the creative process—from the planning phase to the choice of technologies and programming languages—to bring MidasCV to life?

HM: I started with what was familiar to me: Angular for the template. At the time, the interface wasn’t yet complete, but the template directly rendered a local JSON file, extracted using a Python script. I used Selenium to fetch data directly from LinkedIn. When I decided to turn it into a real product, I adapted the template to the “JSON Resume Schema”, built the backend with Firebase, and experimented with various methods for extracting data from LinkedIn, until I ended up with a browser extension, which I consider the only truly GDPR-compliant solution.

DR: Can you illustrate the architecture of MidasCV and explain how it guarantees efficiency and scalability?

HM: From a technological standpoint, I chose Firebase because it is a “backend as a service” designed to scale indefinitely, developed by some of the best engineers in the world. The costs are higher compared to an on-premise solution, but the savings in infrastructure development make the payback period long enough that we don’t even have to think about it at this stage.

In the future, we could always migrate to a system on Kubernetes, hosting it on all major providers or even on-premise. The same goes for AI: if provider costs become too high compared to a local solution, we could develop our own model or apply a fair usage or credit policy. The approach is not to solve problems that do not yet exist. Current technologies can handle growth up to a million in revenue without difficulty.

From an operational perspective, using modern technologies like Angular and an optimized code structure ensures horizontal scalability of the team, should it be necessary to speed up iterations. Moreover, we already have legal and accounting infrastructures in place to manage everything. I therefore believe we are well-equipped in every aspect, while remaining aware that new needs will emerge as the product evolves.

DR: Which algorithms or artificial intelligence models have you integrated into MidasCV? Do you use generative models based on fine-tuning custom solutions or standard implementations like ChatGPT?

HM: At the moment I use external providers and have found that the best cost-benefit ratio is achieved with advanced techniques such as chain-of-thought, retrieval-augmented generation, and prompt engineering, rather than fine-tuning.

Fine-tuning is expensive: if you don’t have the proper data and resources, it can even worsen the results. In the case of MidasCV, it turned out not to be necessary. Even OpenAI advises against fine-tuning, as in most cases it is not the best solution. Sure, it can make you feel like a true AI researcher, but it is fundamental to make rational, data-driven decisions to obtain the best results.

DR: What measures have you taken to guarantee the security of users’ data?

HM: The data is stored on Firestore and protected by access rules based on the user. Accounts are managed via Firebase Auth, accepting exclusively OAuth.

We therefore rely on the advanced security systems of providers like Facebook and Google, which apply additional controls—such as two-factor authentication—in case of suspicious activity.

DR: What have been the most complex technical challenges you encountered during development—including the so-called “hallucinations” of the model—and how did you solve them?

HM: AI errors have been managed in two ways. The first is the upstream work through building an agent composed of several prompts, steps, and advanced techniques. The second level is the design that keeps the human in the loop: if the user detects errors, they can correct them manually before downloading or even regenerating the entire CV.

Another challenge was creating a GDPR-compliant solution for extracting data from LinkedIn. The solution we found was to create a browser extension, which proved extremely effective for a multitude of use cases and perfectly in line with our philosophy: empowering the user with automation, not replacing them.

DR: How did you design the user interface to guarantee an intuitive and accessible experience?

HM: Step by step. I design a feature and test it personally as the first user. If I notice friction or complex steps, I look for solutions.

A significant example concerns the installation of the extension. Without a tutorial or instructions, many users installed it but did not know how to proceed. The solution was simple but effective: once the extension is installed, a new tab on LinkedIn automatically opens. If the user is logged in, they see a screen to authenticate on MidasCV, then one to start the free trial, and finally the promised result with evaluations and a personalized CV directly in the list of available positions.

DR: How has user feedback contributed to improving and updating MidasCV?

HM: Little, because they do not provide it. We barely manage to make hypotheses based on their behavior.

For example, I noticed that no one wanted to download the first version of the extension to import data. So I introduced a feature that did it automatically, even if partially, by including a simple link. This generated more activity from users, but in the end I removed it because it was not GDPR compliant, had a higher cost compared to the LLM, and did not bring me closer to the true goal: monetizing the right target, namely those who sit at their PC to send applications and have no problem installing an extension, being technical users.

DR: Have you collaborated with or integrated MidasCV with other digital services? What synergies have had a significant impact on the project?

HM: Not yet. I have proposed to a couple of companies in the recruiting sector to explore an affiliate or white label model, but I still need to send many emails.

DR: How do you measure the success of the platform and which indicators do you consider fundamental for evaluating the effectiveness of matching candidates’ skills with companies’ needs?

HM: Midas is an agent with multiple personalities and one of them believes it is a recruiter. It selects candidates exactly as a human would: it reads the CV, evaluates its completeness, and identifies any gaps in the career path.

The only aspect it currently does not evaluate is the salary offer, because that information is not available. However, in the future we might add new variables to this "mental process," making the service even more complete and sophisticated.

DR: Which stakeholders do you think MidasCV serves most: companies, candidates, or both? And in what way does it support each of these groups?

HM: I believe it is more useful to recruiting companies because it addresses problems they face on a daily basis, while for candidates it is a need that arises only a few times a year for freelancers, or once every two years for employees. For companies, MidasCV represents a tool to present their candidates in the best possible light.

DR: How do you evaluate the effectiveness of MidasCV in improving selection processes and in highlighting the real merit of candidates?

HM: I think it is the best automated tool available on the market—otherwise, I wouldn’t have bothered creating it. It encapsulates all the knowledge I have distilled over the years to learn how to enhance skills and improve communication with recruiters. For candidates, it helps reduce off-target applications and provides recruiters with clear and concise information, facilitating their work. For companies, it helps automatically filter out clearly unsuitable profiles and assists recruiters in presenting the candidates to client companies in the best possible way.

DR: Have you had any previous experience with artificial intelligence, perhaps in different fields? If so, how did those experiences influence the development of MidasCV?

HM: No, it was literally my first approach with LLMs.

DR: Looking at the world of work, what is your personal vision of how the pervasiveness of AI might change our way of working, both in positive and negative terms?

HM: I believe that at most the number of professionals required to meet market demand might be reduced, but I see AI as an empowerment tool, not a replacement. However, just in case, I bought a house and started growing vegetables, in case I get replaced. I could sum up my attitude as cautious optimism.

DR: Do you think that solutions like MidasCV, by automating partly creative processes, could in the long term diminish people’s creative capacities?

HM: Not today; at most they can accelerate and improve creative processes. But one day we might witness the birth of AGI and discover that we are no longer at the top of the food chain. Or we might realize that there is something in human beings that is irreproducible by a machine. That day will come, and the only wise thing we can do is accept it, whatever the answer may be.

DR: Looking back on the development experience of MidasCV, what have you learned? Do you already have future updates in the works? What advice would you give to someone who wants to embark on a similar project?

HM: At the moment I am focusing mainly on communication. I have reached a satisfactory qualitative level and now I am trying to involve 10 users. For this reason I am working on content, advertising, and sales to understand if and how the product appeals.

For anyone wanting to launch a similar project, I have only one piece of advice: go for it. There is no lesson I can pass on—you can only learn on your own. You are unique, your experience is unique, as will be the tailored solutions for you. The other advice is to never buy courses or consultancy: those who know how, do; those who don’t, teach.

DR: Open question... Tell us whatever you prefer: something about yourself, your vision of the world, your aspirations, or simply what you wish to share with our readers.

HM: Thank you for the opportunity to present my project to your audience and for your curiosity. The last thing I would like to leave for the readers is a message of hope and confidence in their own abilities. I believe that we will live in a better world if everyone tries every day to exercise their virtues, without living in fear, without letting others think for them. It is not necessary to found a company or develop an artificial intelligence product, but I can say with certainty that if you give your best in something you are passionate about, you will live a more fulfilling life—whatever that means for you.

 

Thank you for your time.

At Turtle’s AI we love spreading good projects, like yours.

We emphasize that this article does not contain any sponsorship and has not received any economic contribution. I proposed this interview to Henrique completely free of charge, as is the tradition of Turtle’s AI, because his project particularly impressed me.