Miguel Fierro
4x AI Founder • xMicrosoft • I help people bridge the gap from learning AI theory to getting AI results using my method “Reverse Learning”
United Kingdom
79K followers
500+ connections
About
You see AI everywhere — in the news, in your company, in every job posting. And there's a voice in your head saying: "I'm falling behind."
You've tried to keep up. You use ChatGPT. You've taken courses, maybe have a certificate. And yet when someone asks "what can you do with AI?", you don't have an answer.
It's not your fault. The entire AI education industry is designed to sell you knowledge, not to help you build proof.
The professionals who are actually getting ahead with AI (hired at top tech companies, getting promoted, leading projects), have one thing in common: they've built something real, a result that can be verified.
I'm Miguel Fierro. I've spent 10 years at Microsoft building production AI. I'm the creator of Recommenders (20K+ GitHub stars, used by Fortune 500s) and 4x AI founder.
After 10 years at Microsoft, I noticed a pattern nobody talks about: companies don't want people who know AI theory, they want people who build useful AI and leaders who can drive that transformation.
At Microsoft, I watched this pattern again and again. The developer who shipped one real AI project got the offer over candidates with 10 courses on their resume. The business leader who automated one workflow got the promotion over colleagues who could only talk about AI.
The gap is never knowledge. It's proof.
- Developers don't need another tutorial, they need a deployed project, a clean GitHub repo, and a README that tells a story a recruiter can follow in 30 seconds.
- Business professionals don't need to learn to code, they need one working AI workflow that saves time or makes money, and a case study that proves it.
Now I help people bridge their careers to AI, by using a method called Reverse Learning: instead of months of theory first, you start by building and learn what you need along the way.
Some people I helped:
- Daniel Labbe: from Solution Architect to AI Engineer at Microsoft.
- Nikolaos Zavitsanos: from Software Engineer to AI Engineer at AWS.
- Paolo Colecchia: from Software Engineer to AI Engineer at Microsoft.
- Tanuja Bhamidipati: from Software Engineer to AI Solution Architect at Microsoft.
- ... and hundreds more.
If you're ready to stop stop collecting courses and start building proof, send me a DM, and I'll see how I can help you.
Activity
79K followers
-
Miguel Fierro shared thisOne big problem with agents: They forget everything between sessions. Solution? Memory engineering. Most agents wake up blank. The conversation ends and the context dies with it. A new discipline is emerging to fix it: memory engineering. 𝗧𝗵𝗲 𝗳𝗶𝘃𝗲-𝘀𝘁𝗮𝗴𝗲 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲. → Capture: record what happened this session. → Consolidation: a scoring layer decides what matters beyond it. → Retrieval: bring back only what the task needs. → Reconciliation: contradictions get flagged, never silently overwritten. → Decay: memory that is never touched fades out on its own. 𝗧𝗵𝗲 𝗸𝗲𝘆 𝗶𝗻𝘀𝗶𝗴𝗵𝘁. Storing everything is not memory. It is bloat. Memory is deciding what survives. The scoring layer is the product. Context engineering was the skill of 2025. Memory engineering is next.
-
Miguel Fierro reposted thisMiguel Fierro reposted thisCada cliente ve una marca distinta. Y tú no lo sabes. Durante décadas, el marketing ha hablado a audiencias: segmentos, targets, medias. Eso se ha terminado. Los algoritmos que deciden qué ves en #Instagram, #TikTok o #Amazon ya tratan a cada persona como un público de uno. Aprenden de cada cliente y le muestran una versión distinta de cada marca. El 21 de Octubre, Miguel Fierro (Founder, RecoMind) abre esta conversación en la sesión inaugural de AI Minds by Mainkore. Nos ayudará a entender qué hay detrás de esa tecnología y qué significa para quienes toman decisiones de marketing y negocio. La pregunta de fondo con la que arrancamos: Si cada cliente está viendo una versión distinta de tu marca, ¿sigues teniendo realmente una sola marca? 21 de Octubre· en #ForbesHouse. Plazas limitadas. [https://lnkd.in/e2nZHfmi] #AIMindsByMainkore
-
Miguel Fierro shared thisAnthropic pays $3,850 a week. No ML experience needed. No PhD either. It is called the Anthropic Fellows Program. 4 months of full time research, paid weekly. Plus around $15K a month for compute. 𝗪𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗮𝘀𝗸 𝗳𝗼𝗿. → Fluent Python. → Full time availability. → Work authorization in the US, UK or Canada. 𝗪𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗱𝗼 𝗻𝗼𝘁 𝗮𝘀𝗸 𝗳𝗼𝗿. → A PhD. → A publication record. → Years of ML on your CV. An Anthropic researcher proposes project ideas. You pick one and work on it with them for 4 months. Real research, not another course. 𝗧𝗵𝗲 𝗻𝘂𝗺𝗯𝗲𝗿𝘀 𝗳𝗿𝗼𝗺 𝗽𝗮𝘀𝘁 𝗰𝗼𝗵𝗼𝗿𝘁𝘀. → In one earlier cohort, over 80% produced a paper. → 25% to 50% received a full time offer. One catch. There is no visa sponsorship. Check that rule before you spend a night on the application. The door is open and it says Python. Comment below if you want my ideas on the application.
-
Miguel Fierro shared thisI asked ChatGPT what you admire about humans. Its final answer was: "So if I had to choose just one thing, it would be this: your ability to imagine a better future and then work, however imperfectly, toward making it real." Let's keep building fellow humans!
-
Miguel Fierro shared thisThis is the reason I dedicated my life to AI and robotics. AI is the best way to make scientific and engineering progress. 🚀 OpenAI used an unreleased model, more capable than GPT-6 Astra, to solve the Navier-Stokes problem, a 90-year-old mathematical mystery behind how liquids and gases flow. Key Facts: * The Math: Navier-Stokes equations power weather forecasting and aerodynamics. OpenAI proved that smooth fluid motion can develop a "finite-time singularity" (infinite speed under specific vortex conditions). * How AI Did It: ~10,000 autonomous agents ran for 88 hours, generating a 165-page proof verified line-by-line via Lean code. * The $1,000,000 Prize: Clay Mathematics Institute offers $1M for solving Millennium Problems (only one other has ever been solved). OpenAI stated they will not claim the money. I studied the NS equations when I was studying engineering. I remember I found them super difficult to understand. If this doesn't make you start learning IA, I don't know what else to say.
-
Miguel Fierro shared thisOne of the best ways to reduce the AI bill: LLM model routing. LLMRouter is an open source project that picks the model per request. Not per project. 𝗪𝗵𝗮𝘁 𝗶𝘁 𝗱𝗼𝗲𝘀. → 16+ routing strategies: KNN, SVM, MLP, graph-based, hybrid. → Multi-turn routing that keeps the conversation in context. → Multimodal routing for text, vision, and time series. → Routing memory that reuses past query-to-model decisions. Works with OpenAI, Anthropic, NVIDIA, and local models. 𝗧𝗵𝗲 𝗽𝗮𝗿𝘁 𝗻𝗼𝗯𝗼𝗱𝘆 𝘄𝗮𝗿𝗻𝘀 𝘆𝗼𝘂 𝗮𝗯𝗼𝘂𝘁. → A router is a model, so it drifts. → Your query mix in August is not your query mix in December. → Version the router. Log every decision. → Alert when the share of cheap-model routes moves. The question stopped being "which LLM is best?". It is "which LLM is best for this request?" Link: https://lnkd.in/eV-jCJ6R Bookmark this post.
-
Miguel Fierro shared thisAI is down, there is a massive outage. OpenAI, Claude and xAI are down. What do you do where there is no AI?
-
Miguel Fierro shared thisFDE interview. The hiring manager: "The agent got worse. Why?" The client flagged it this week. "I would check the logs and tweak the prompt" ends the interview. You cannot explain a regression you never measured. The strategy for running AI agents in production is not a bigger model. It is an eval stack. 11 things your LLM app needs before the next deploy. 𝗤𝘂𝗮𝗹𝗶𝘁𝘆. → An eval harness that runs before every deploy. → A golden dataset with 100+ test cases. → LLM-as-a-judge with calibrated scoring. → Regression detection that blocks bad deploys. 𝗦𝗮𝗳𝗲𝘁𝘆. → Hallucination rate with factuality checks. → Grounding quality with citation validation. → Safety evals for toxic or biased outputs. 𝗢𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻𝘀. → Latency percentiles: p50, p95, p99. → Cost-quality curves for model selection. → Shadow deployments and canary releases with automatic rollback. → Dashboards with quality trends and alerts when quality drops. The pattern behind all 11 items. Treat AI quality like uptime. Not a vibe. A number you watch. What would you add to this list?
-
Miguel Fierro shared thisHere's a question I get a lot. Where do I start my career? PG and LeCun answered... Someone asked Paul Graham what he would do if he were 17 today. Yann LeCun answered the same question. Their advice points in opposite directions. 𝗜𝗳 𝘆𝗼𝘂 𝗯𝘂𝗶𝗹𝗱 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝘀, 𝗳𝗼𝗹𝗹𝗼𝘄 𝗚𝗿𝗮𝗵𝗮𝗺. → Learn to build LLMs from scratch. → Train the most powerful one your hardware allows. → Borrowed hardware beats hardware you wish you had. 𝗜𝗳 𝘆𝗼𝘂 𝗹𝗼𝘃𝗲 𝘀𝗰𝗶𝗲𝗻𝗰𝗲, 𝗳𝗼𝗹𝗹𝗼𝘄 𝗟𝗲𝗖𝘂𝗻. → Ask why an LLM writes your essay but cannot clean your bedroom. → Study the topics that could solve that. → Design the architectures that come after LLMs. Graham bets on the technology that already works. LeCun bets on the problem nobody has cracked. Neither said wait until you are ready. The bottom line. Whatever you prefer. Start today. Let me know how I can help.
-
Miguel Fierro liked thisMiguel Fierro liked thisI saw this Content Provenance Standards Program Manager job pop up: https://lnkd.in/eG9dDp8C. Looks pretty interesting in an important and constantly evolving area.| Microsoft Careers| Microsoft Careers
-
Miguel Fierro liked thisMiguel Fierro liked this¿Acabará la IA con las croquetas? Cuando Ricardo Tayar me invitó a #ConversionTalks de Flat 101, le dije que no. -Pero que yo no vendo nada, Ricardo.- Pero insistió. -Tú cuenta tu visión de IA y lo que haces con ella- Así que dije... -¡Venga, cuenta conmigo!- Estaré rodeado de cracks del CRO, conversión y negocio digital. Yo, que solo hago frikadas! 🤷♂️ Tienen aforo completo desde hace semanas (en el próximo tendrán que habilitar reventa). Pero si estás por allí, párame y nos damos un abrazo... en eso sí soy muy bueno! PS.- En realidad hablaré de si la IA acabará con la creatividad. Pero me toca justo antes de la comida. Seré lo único que se interponga entre la audiencia y las croquetas. Así que voy a tener que ser muy creativo para que les merezca la pena la espera. ¡A ver qué se me ocurre!
-
Miguel Fierro liked thisMiguel Fierro liked this"We don't need another AI feature. We need an AI architecture." As an engineer, that's the best thing a customer can say. Adding a chatbot onto an HR system is easy. Building one that actually learns is not. Here's what we're building with Atento: 🧱 An AI-native layer on top of SAP SuccessFactors, working with their data where it already lives. 🤝 Agents that share context across the whole employee lifecycle, not isolated tools in silos. 🔁 Learning loops: every hire, onboarding and career move becomes a signal that improves the next decision. The hard part isn't the model. It's turning years of people data into something agents can reason over in seconds. That's the problem our team obsesses over, and seeing it come to life at Atento is a big moment for us. Welcome, Atento!
-
Miguel Fierro liked thisMiguel Fierro liked thisRecently I posted a short note about the quiet layoff at Microsoft. It's soft, and it's happening. The company is shifting to a new reality. And I'm now part of those numbers myself: I left Microsoft. On posts like this, people usually thank the company, the people they worked with, and the amazing products they built. My note is different. I appreciate the time I spent there. I learned how huge corporations operate, how politics works, and what it means to be part of a system (in some cases, even a hostage of it). I collected valuable material about management. I had read books about good and bad management and couldn't believe some of those cases really existed. They do! And now my experience is richer for it. I worked with skilled, brilliant people who deserved their positions. But there were few of them, and that makes my connections with them even more valuable to me. I also appreciate the company merch! I stocked up on good-quality stuff: a nice jacket I love to wear, a decent hoodie, and a backpack I've been using since my first day at Microsoft. I'd put its quality on par with Tochka's :) I'm proud of my contribution. People grew in the team I led, got promoted, and achieved their goals, even when that meant moving from my team to another one. Helping people pursue their goals and be happier is one of my values. Sometimes it hurts. But people are the most important part of any team. And that also played against me. There were political games around me, and I never learned to play them. I didn't play even when I knew the outcome. Between being part of the system and following my values, my values won and politics lost. But once again, there's no positive or negative experience, only learning. I learned. I know :) Thank you for the chance! P.S. Yep, it's Apple, not Windows
-
Miguel Fierro liked thisConsequential work by the team!Miguel Fierro liked thisLast year, Google DeepMind Project Astra showed Demis Hassabis' vision of a universal AI assistant that can be a visual interpreter & see the world around you. Today, we took it from research prototype to all Gemini App users, at scale. Guided Vision is now live in Gemini Live for all Android users globally, in 50+ languages. For the past year, the blind, low-vision & aging community has been our partner, showing us what "helpful" really means. Safety was designed in with the community from the very start. Turning a research demo into something people can rely on in their own homes has been the most meaningful work I've done. With gratitude to: Aira whose real-world data (tens of thousands of hours!) made the model better National Association for the Blind, Delhi SG Enable Fundação Dorina Nowill para Cegos & many more trusted partners, for your feedback & honesty The Deepmind Impact Accelerator team Alap Bharadwaj Doruk Caner ; Gemini for Aging Seniors Motoki Sano The amazing Google teams who built this together: Scott Adams Alistair Muldal Angana Ghosh Maura Lynch Perrin Anto Harshil Ved Tara Thomas And my fellow Gemini Live partners in crime Aditya Barua Sookyoung Kwak Meena Kaushik Michaela L. under Hyunjeong (HJ) Choe & Robert Caldwell's leadership. I couldn't have asked for a better crew. This is just the beginning, we'll be expanding access to more platforms and form factors. If someone you know is blind, has low vision, or simply struggles with fine print, I'd be honored to hear what they think. Read my post on The Keyword: https://lnkd.in/g6wQKp_G
-
Miguel Fierro liked thisMiguel Fierro liked thisPlease help!!! 🚨 We’re looking for a Senior Software Developer. 💰 $150K–$190K 📈 Stocks 🏥 Private medical insurance 💻 Hybrid Just one small requirement: 10+ years of experience with Claude Code. (Claude Code launched in 2025.) But since apparently nobody in the HR team noticed that… I assume lying on your CV is also part of the job requirements too. 😂
-
Miguel Fierro liked thisMiguel Fierro liked this⚙️ Ask any data team where their time really goes, and the answer is rarely "insight". It is pipelines, preparation, and fixing what broke overnight. That is why the 𝗮𝗴𝗲𝗻𝘁𝗶𝗰 𝗱𝗮𝘁𝗮 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 news from FabCon Europe caught my eye. As CIO.com reports, Microsoft Fabric is introducing new capabilities, building on the Osmos acquisition, where: 🎯 Engineers define an outcome and set the boundaries 🤖 An agent autonomously executes the work ✅ It validates the results and refines its own approach 🗂️ Raw data becomes analytics and AI-ready assets in OneLake What I like is the 𝗱𝗶𝘃𝗶𝘀𝗶𝗼𝗻 𝗼𝗳 𝗹𝗮𝗯𝗼𝘂𝗿. People decide what good looks like and where the limits are. The agent handles the repetitive execution and checks its own output. That is the right model for agents in any serious enterprise setting. The same article describes building entire applications end-to-end in Fabric, fully hosted, managed and governed. 𝗙𝗲𝘄𝗲𝗿 𝗺𝗼𝘃𝗶𝗻𝗴 𝗽𝗮𝗿𝘁𝘀 𝗮𝗻𝗱 𝗹𝗲𝘀𝘀 𝗴𝗹𝘂𝗲 𝗰𝗼𝗱𝗲 is exactly what customers keep asking me for. 🔗 Read it here: https://lnkd.in/eNJ7fYZb 👥 Every day I get to see our Europe South Solution Engineering team help customers modernise their data estates this way: André Cigarro Matos Álvaro Segura Manzanares David Mangas Nuñez Pier Luigi Croce Giovanni Gallileo Liliana Pasceri Nikolaos Antoniou Gokhan Tufan Ido Katz Arik Bidny. 🙏 Thank you as well to Ali Tahmasebi and Jasmin Zilkens for the leadership and partnership across EMEA. #MicrosoftFabric #AgenticAI #DataEngineering #OneLake #Microsoft
Experience
Education
-
Massachusetts Institute of Technology - Sloan School of Management
Business Administration, Management and Operations
-
-
Kings College London
PhD in robotics Robotics Excellent Cum Laude and Outstanding Thesis Award
-
Activities and Societies: Humanoid robots, mobile robots, computer vision, learning from demonstration, skill innovation, intelligent systems, social robots, path planning, kinematics and dynamics
PhD at Universidad Carlos III de Madrid in collaboration with King's College of London. I had over supervisor from Carlos III and another from King's.
My thesis covers non-linear control, machine learning, postural planning and control of humanoid robots. Development of computer vision algorithms, objects' localization and environment mapping. Trajectory optimization and path planning using genetic algorithms and neural networks.
I was awarded the Outstanding Thesis Award of UC3M.…PhD at Universidad Carlos III de Madrid in collaboration with King's College of London. I had over supervisor from Carlos III and another from King's.
My thesis covers non-linear control, machine learning, postural planning and control of humanoid robots. Development of computer vision algorithms, objects' localization and environment mapping. Trajectory optimization and path planning using genetic algorithms and neural networks.
I was awarded the Outstanding Thesis Award of UC3M. A list of my publications can be found in https://github.com/miguelgfierro/publications. -
Universidad Carlos III de Madrid
Master in Robotics and Automation
Activities and Societies: Humanoid robots, Mobile robots, artificial vision, intelligent systems, social robots, spacial robots, path planning, kinematics and dynamics
First class honours (top 1%). Degree finished in half the time (9 months) where the normal duration is 1 year and a half.
-
Universidad Carlos III de Madrid
Industrial Engineer Electrical Enginneering
-
Activities and Societies: Physics, analog and digital electronics, control and robotics. Fluids, electromagnetism, partial and differential equations, mechanics, structure analysis, management, marketing, projects management.
First class honours (top 5%)
Publications
-
Link to all scientific and technical publications
-
See publicationIn this github repo https://github.com/miguelgfierro/publications you will find some of the publications that I have done during my career.
Patents
-
ssm library: Image recognition and similarity for the fashion industry
Issued ES M-008319/2014
The solution is a visual search engine adapted to fashion products. The algorithm, named
Fashion Recognition, is able to recognize a garment in a picture and find a set of garments with a similar
appearance. The methodology is based on a pipeline consisting of segmentation, feature extraction, feature hashing and feature comparison.
Honors & Awards
-
Top 11 Creator on LinkedIn 2025
Favikon
Included among the top 20 Spanish creators on LinkedIn
-
Top Voice on LinkedIn 2025
LinkedIn
Given the Top Voice award for 2025.
-
Leading voice in AI 2024
SmartDev
Named one of the leading voices in AI together with leaders in the field like Demis Hassabis (Nobel laureate), Yann LeCun (Turing award winner), Andrew Ng (founder of Coursera), Thomas Wolf (founder of Huggingface), and more.
-
Teaching Excelence Award at IE University
IE University
Teaching Excellence Award for the course on Chatbots and Recommendation Systems.
-
Top Voice on LinkedIn 2024
LinkedIn
Given the Top Voice award for 2024.
-
2nd top AI creator in Spain on LinkedIn
Favikon
I have been recognized as the 2nd top AI creator in Spain on LinkedIn. I'm also among the top 200 creators on LinkedIn in Spain.
-
Top AI Voice on LinkedIn 2023
LinkedIn
Given the Top AI Voice award for 2023.
-
Data Scientist of the Year
Tech Leaders Awards UK
The Tech Leaders Awards recognise UK’s top tech leaders, innovators and disruptors, who have demonstrated value from the application of technology and drived disruptive innovation in large organizations.
-
University Carlos III Excellence Alumni Award
Universidad Carlos III de Madrid
The Excellence Alumni Award is the highest honor bestowed by the University Carlos III on alumni. The award recognizes individuals who have demonstrated excellence in their professional careers.
-
Silver medal at Passenger Screening Algorithm Challenge
Kaggle
Silver medal in the Kaggle competition of airport passenger screening (the competition with the highest prize so far $1.5M)
-
Outstanding Thesis Award
Universidad Carlos III of Madrid
Outstanding Thesis Award in the Doctorate program of Electrical Engineering, Electronics and Automation.
This award is granted by the University to the best PhD thesis of the year 2015. -
Finalist of BBVA OpenTalent
BBVA
Finalist of the international startup competition OpenTalent among more than 900 projects in the category of technology and innovation.
-
First prize in the 6th Enterpise Challenge at Universidad Carlos III of Madrid
Universidad Carlos III de Madrid
First position at the University Contest, award given by a Jury formed by Private Investors, Business School members and Business Angels.
-
First Prize 2020 Startup Madrid
Olympic Games Organization for Madrid2020
We were selected to represent the olympic game spirit in the candidature of Madrid 2020 for the Olympic Games. The prize was provided by the City Council of Madrid.
-
Finalist of Top Seeds Lab
Top Seeds Lab
We were selected as finalist to join Top Seeds Lab accelerator.
-
First Prize Madrid Emprende
Universidad Autónoma de Madrid
We won the first prize Madrid Emprende X Premio CIADE al Emprendedor Universitario of Universidad Autónoma de Madrid among 200 projects.
-
Finalist RedEmprendia challenge: Ideas en Acción
RedEmprendia
International competition for Spanish, Portuguese and South American spin-offs. Finalist among more than 300 projects.
-
Winner of the VI RobotChallenge 2009 European Robotics Championship
-
Robotic competition of minihumanoids hosted in Vienna. We achieved the first position in the sumo competition and second position in the running competition.
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More