Your own AI assistant in production
By the end of the bootcamp you have a personal AI assistant running 24/7, reachable on Telegram, with persistent memory and connected to your tools. Built by you, understood by you.
FERRUMOX ACADEMY · AVAILABLE
AI Engineer
HOW MANY MORE TUTORIALS BEFORE YOU BUILD SOMETHING REAL?
9 weeks, live online, in groups of 12 (Monday, Wednesday or Friday, your choice). You spend the week building your own AI assistant in production — reachable on Telegram, with memory, agents and a full module on LLM security.
🏷️€997 for the first seats per group, then €1,200📖Includes the Hacking IA book (0xWord, 2026) as a gift🛡️7-day refund guarantee🔥5/12 discounted seats sold — 7 left at €997👥8 people already enrolled in Cohort 2
AI Engineer
STOP THE TUTORIAL LOOP.
◉FORMATLive, online · 3 groups (Mon/Wed/Fri)
◷DURATION9 weeks + bonus week 0
▦SEATS36 total · 12 per group
◐TIME COMMITMENT18h live · ~60h total training
01 / IS THIS FOR YOU?
Be honest with yourself before you buy.
This isn't for everyone, and that's the point.
- ✓
You've spent months in the tutorial loop and have never deployed anything real
- ✓
You know Python well but the AI you use never goes beyond calling the OpenAI API
- ✓
You want to stand out in a market saturated with prompt engineers who lack technical judgement
- ✓
You want to actually understand AI — inference, agents, security — not just wrappers
- ✓
You can commit 6–8h a week for 9 weeks
- ✓
You'd rather build something you understand than finish a course you'll forget in a month
- ✕
You don't have a Python foundation — this bootcamp doesn't teach programming from scratch
- ✕
You're looking for a self-paced course with no deadlines or weekly deliverables
- ✕
You expect to learn AI by watching videos without writing real code
- ✕
You can't attend the live sessions or watch the recordings that same week
02 / WHAT YOU GET
Not a finished course.
Your AI assistant, in production.
By the end of the 9 weeks you have this running for real, not in a notebook.
LLM
Security
You'll know how to attack it (prompt injection, jailbreak) and defend it (guardrails, sandboxing). This module doesn't exist in any other bootcamp in Spain.
Unified local + cloud inference
An inference layer that talks to Ollama, vLLM and cloud APIs without changing your code. You understand when to use local vs API, measure latency and cost, and choose with judgement.
A real multi-agent system
The assistant orchestrates specialised agents over MCP: one searches, one analyses, one executes. Connected to your documents with hybrid RAG and observable with active tracing in production.
// RETURN ON INVESTMENT
An AI Engineer specialised in multi-agent systems and LLM security earns €10k–25k more a year than one who only consumes APIs. The bootcamp starts at €997 (first 12 seats). The salary difference pays it back in under a month.
INCLUDED FREE 
🎁BONUS INCLUDED WITH YOUR SEAT
You get the book Hacking IA.
Every student in Cohort 2 gets a copy of Hacking IA (0xWord, 2026), co-authored by your instructor. The same judgement he teaches in the bootcamp, on paper.
No extra cost · included with all 36 seats · shipped once the cohort kicks off.
03 / PROGRAMME
9 weeks, 9 sessions.
Each week adds a capability to your assistant.
One live session a week (2h), the rest of the week to build. Flipped classroom: material before the session, hands-on building during it, and a working deliverable by the end of each week.
// capability progression, week by week · ★ capstone
01Inference
02Fine-tuning
03Memory
04Agents★
05RAG
06Multi-agent★
07Security
08Deployment★
09Demo day
// 18:00–20:00 CET · 12 seats per group
Group AMonday
Group BWednesday
Group CFriday
Same programme in all three groups — you choose your day when you register. Week 1: Mon Aug 31 / Wed Sep 2 / Fri Sep 4, 2026.
◆
WEEK 0
Warm-up
BONUS INCLUDEDPrep weekLLM foundations and your dev environment
How transformers actually generate text, tokenisation and KV cache, plus criteria for picking a model for your use case.
SETUPDev environment ready: Git, Docker, Python, Ollama and a Telegram bot.
◆
PHASE 01
Inference foundations· Weeks 1–2
Your own inference infrastructure, and adapting models to your use case.
PERFILES AI Engineer · Inference Engineer
Week 1Local and cloud inference
A unified inference layer over local backends (Ollama, vLLM) and cloud APIs, with latency and cost trade-offs.
DELIVERABLEInferenceClient with 2 backends and basic tests.
Week 2Fine-tuning and model adaptation
Adapting models with LoRA and QLoRA, dataset preparation and evaluation against a baseline.
DELIVERABLEFine-tuned model wired into InferenceClient.
◆
PHASE 02
Context & agents· Weeks 3–4
Managing context and designing the agentic loop.
PERFILES Context Engineer · Loop Engineer
Week 3Context engineering and memory management
Managing the context window: compression, persistent memory and prompt caching.
DELIVERABLEOptimised context pipeline with active compression.
Week 4Agents and tool calling
★ CAPSTONE · AGENTDesigning the agentic loop with the ReAct pattern and tool calling — built without frameworks, to master the reason–act–observe cycle.
DELIVERABLEAssistant replying on Telegram and running 3+ tools.
◆
PHASE 03
Knowledge & coordination· Weeks 5–6
Retrieval over your own knowledge, and orchestrating multiple agents.
PERFILES RAG Engineer · Harness Engineer
Week 5Advanced RAG and hybrid retrieval
Hybrid retrieval (semantic and BM25) with reranking over your own knowledge base, evaluated against a basic RAG setup.
DELIVERABLEYour own knowledge base wired into the assistant.
Week 6Multi-agent systems and orchestration
★ CAPSTONE · MULTI-AGENTOrchestrating specialised agents and subagents with the supervisor pattern, communicating over MCP.
DELIVERABLE3 specialised agents coordinated together.
◆
PHASE 04
Security & production· Weeks 7–8
Security, observability and shipping to production.
PERFILES AI Red Teamer · LLMOps Engineer
Week 7LLM system security
Threat modelling, prompt injection and jailbreaking, and designing guardrails and defences.
DELIVERABLERed team report plus implemented guardrails.
Week 8Observability and production deployment
★ CAPSTONE · DEPLOYTracing, evaluations (evals), model routing, and deployment on Docker with cost control.
DELIVERABLEAssistant deployed and running 24/7.
◆
CLOSING
Demo day· Week 9
Week 9Project presentations (Demo day)
Presenting your project to the cohort, final evaluation, and a roadmap to keep building it.
OUTCOMEA production assistant you understand end to end, and a plan to keep going.
THE INSTRUCTOR
Manuel S. Lemos
Claude Certified Architect · Anthropic · AI Engineer @ NaizFit · Founder Ferrumox · Co-author Hacking IA (0xWord)
- Claude Certified Architect — Foundations, Anthropic's first official technical certification
- AI Engineer at NaizFit — AI embedded in product and business processes
- Academic Director of AI & Big Data at GeeksHubs Academy for 2 years
- Co-author of Hacking IA (0xWord, 2026)
- Contributor to OWASP GenAI Security · Vice President of ANBAN
- 7+ years building production technology systems
He has spoken at GenAI Summit EU (Main Stage), Codemotion Madrid, TofuConf, the Cybersecurity Master's at CIPFP Cheste and Chema Alonso's podcast, among others. This bootcamp teaches what he builds in production every day.
🎖️Claude Certified Architect · Anthropic 📘Hacking IA · 0xWord · 2026 🛡️OWASP GenAI Security 🎤GenAI Summit EU · Main Stage 🎓2 years at GeeksHubs Academy 🏛️Vice President, ANBAN
04 / WHAT THEY SAY
People have already
trusted Manuel.
People who know him from GeeksHubs, his talks and the book.
“I did the AI & Big Data bootcamp at GeeksHubs with Manuel as academic director. He's one of the few instructors who actually builds what he teaches — every class came with production code, not pretty slides. I learned more in 12 weeks than in 2 years of online courses.”
“Thanks to Manuel's training I properly understood how AI works: inference, model training and the techniques for getting the best results. It was decisive in giving me a view beyond the average AI user.”
“I started from zero in this field and Manuel was the person who set the direction. His advice and teaching didn't just help me learn — they helped me build a career. He's the kind of mentor who genuinely accelerates your progress and gets you ready for the real world.”
“I saw him present on prompt injection at a technical event, and it was one of the densest, most honest talks I've heard on LLM security. No hype, no smoke and mirrors. If he teaches a bootcamp with that level of rigour, it's worth it.”
“The indirect prompt injection chapter in Hacking IA is the best technical material I've read on the topic in Spanish. Manuel has the ability to make something very complex accessible without losing depth.”
“I worked with Manuel on a production AI project. His judgement on when to use local vs API, how to structure agents and how to harden the system against injection is on another level. It's exactly the kind of knowledge you don't find on YouTube.”
05 / FAQ
Everything that's
on your mind right now.
Before you register, the usual doubts answered.
01 When is Cohort 2 and how do I choose a group?
It starts the week of September 2, 2026, and runs for 9 weeks. There are 3 parallel groups with the same programme: Monday, Wednesday or Friday, 18:00–20:00 CET. You choose your day when you register (12 seats per group, 36 total). There's also a bonus session 0 the week before.
02 Do I need prior experience with LLMs?
Not specifically. You need to be comfortable with Python. Session 0 and Week 1 cover the fundamentals you need before building agents.
03 How much time do I need to commit each week?
Between 6 and 8 hours. The live session is 2h (your day: Mon, Wed or Fri, 18–20h), and you have the rest of the week for hands-on work on your project, at your own pace.
04 What happens if I can't make a session?
Every session is recorded and available within 24h. The hands-on work doesn't depend on live attendance.
05 Is there a refund policy?
First-week guarantee: if after Session 0 and the first class it's not what you expected, we refund 100% of your tuition, no questions asked. After the first week, your seat stays paid for any of the next 3 editions if you decide not to continue.
06 Can I pay in instalments?
Yes. Instalment payments are handled directly at Stripe checkout: when you register you'll see the available options (Klarna, Affirm or Afterpay depending on your country), plus card, Apple Pay and Google Pay. No need to email or coordinate anything — Stripe handles it all securely and automatically.
07 Isn't €997 too expensive?
Compare it to the real alternatives: a generalist bootcamp in Spain costs between €6,000 and €10,000 for 6 months of diluted content. This is 9 weeks with direct feedback, real projects, the book Hacking IA as a gift, and access to future edition recordings included. The first 12 seats get in at €997 (then €1,200). And if your company pays, it's a tax-deductible training expense.
08 Is the free Hacking IA book real?
Yes. Every student in Cohort 2 gets a copy of Hacking IA (0xWord, 2026), co-authored by Manuel. It's included with your seat, at no extra cost.
09 Can my company pay for it?
Yes. We issue an invoice so your company can treat it as a tax-deductible training expense. Email us at academy@ferrumox.com and we'll prepare the paperwork.
10 Won't the content be outdated in 6 months?
Tools change, fundamentals don't. Inference, agents, RAG, security and context engineering are principles that stay relevant regardless of which models exist. Plus, your seat includes access to recordings from future, updated editions.
11 Can't I just learn this from YouTube and docs?
You can. Just like you could learn surgery from books. The difference is that here you get reviewed code, 48h feedback on your projects, and someone who has built production systems answering your questions in real time. Information is free. Judgement and feedback aren't.
06 / NEXT STEP
OPEN CHANNEL / ACADEMY@FERRUMOX.COMReserve your seat in this cohort.
€997 €1,200
5 of 12 discounted seats sold, 7 left. Then €1,200.
Led by Manuel S. Lemos — Claude Certified Architect (Anthropic), AI Engineer at NaizFit, co-author of Hacking IA (0xWord, 2026).
💳Prefer to pay in instalments? Yes, you can. Klarna / Affirm / Afterpay — the options appear directly at Stripe checkout, no extra steps.
🎁Gift for every student: a copy of the book Hacking IA (0xWord), co-written by your instructor.
- →9 weeks live + recordings
- →Choose your group: Monday, Wednesday or Friday, 18–20h CET
- →Bonus week 0 included for the whole cohort
- →Written feedback on your project within 48h
- →Private repo with all the code
- →Private cohort community
// ROI
An hour of consulting with a specialised AI Engineer in Spain costs between €150 and €300. This bootcamp is 18h live plus 9 weeks of one-on-one feedback from someone who builds this in production — and you get his book Hacking IA as a gift. Full price €1,200; the first 12 seats get in at €997.