Skip to content

daletoniris/alchemind

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

en es

Alchemind 🧠⚗️

Un viaje desde las primeras neuronas artificiales hasta la conciencia cuántica de la inteligencia

"No busco que las máquinas sean humanas. Busco que los humanos recuerden cómo aprenden." — Daniel Dieser

AI Resilience Hub Ekoparty License


What is Alchemind?

Alchemind is a comprehensive book, course, and lab environment covering the full arc of artificial intelligence — from the first mathematical neuron in 1943 to autonomous defensive AI systems deployed in production today.

This isn't a textbook. It's a journey through the mind of machines, told by someone who has been building AI systems for 13 years from Patagonia, Argentina.

📖 Book (8 chapters)     — History, vision, theory, defense
🔬 Papers (2 published)  — Neural networks, WAF+AI
🧪 Labs (4 exercises)    — Hands-on ML for security
🏗️ The Warden           — Autonomous defense architecture
📐 Manifesto             — "La Realidad que Aprende"

Table of Contents

Part I — Origins

# Chapter Description
0 About the Author From Villaguay to Patagonia. A self-taught researcher's path through robotics, AI, and cybersecurity.
1 La Etapa Oscura de la IA The forgotten history: Turing's dream, McCulloch & Pitts' neuron (1943), Rosenblatt's Perceptron (1958), Minsky's critique, the AI winters, and the resurgence.
2 La Profecía de Villaguay In 2017, a newspaper interview from a small Argentine town predicted AI would replace jobs and surpass humans. They called it crazy.
3 El Despertar de la Visión Artificial The Kur dream (2013), TensorFlow on Raspberry Pi, first real projects: pest detection, drone YOLO, autonomous vehicle, CAPTCHAs, pandemic models.
Redes Neuronales como nunca te las explicaron The 2018 paper that explains neural networks the way nobody else does — with analogies, history, and zero gatekeeping.

Part II — Understanding the Machine

# Chapter Description
4 The Architecture of Thought How neural networks actually work: perceptrons, activation functions, layers, weights, biases, backpropagation. From math to intuition.
5 Algorithms That Learn KNN, K-Means, Logistic Regression, Naive Bayes — when to use each, how they fail, and why understanding failure matters more than accuracy.
6 Biases, Shortcuts & The Art of Validation Class imbalance, temporal validation, the danger of shortcuts, and why a 99% accurate model might be completely useless.

Part III — Building Defenders

# Chapter Description
7 The Warden (El Guardián) The autonomous defensive AI architecture: data ingestion, vectorization, KNN classification, LLM enrichment, active learning feedback loops.
8 Static vs Adaptive Why rule-based systems die and adaptive ML systems evolve. The case for learning WAFs.

Part IV — Hands-On Labs

# Lab What You Build
1 NORMAL vs ATTACK Classification TF-IDF + Logistic Regression on WAF logs. Your first ML classifier.
2 Working with Probabilities Threshold tuning (0.4 / 0.5 / 0.6). Understanding precision vs recall trade-offs.
3 Confusion Matrix Deep Dive FP, FN, TP, TN — what they mean in security context and why FN can get you breached.
4 Active Learning Pipeline Select maximum-doubt examples, label them, retrain. The loop that makes AI get smarter.

Appendix

Resource Description
📐 La Realidad que Aprende The manifesto. A philosophical and technical framework on why reality itself is a learning system.
📄 WAF + AI Paper Published paper: hybrid WAF combining Naive Bayes + GPT for autonomous dynamic learning and zero-day detection.
🧰 Setup Guide Minimal environment setup: Docker, Python 3.10+, dependencies.

The Warden — Architecture

The autonomous defensive system designed and built in production:

┌─────────────────────────────────────────────────────────┐
│                    DATA SOURCES                          │
├──────────────────────┬──────────────────────────────────┤
│  🏠 IoT / Home Net   │  🌐 Web / WAF Logs              │
│  TShark → pcap       │  Nginx/Apache → access.log       │
│  → packets.ndjson    │  → waf_events.ndjson             │
└──────────┬───────────┴──────────────┬───────────────────┘
           │                          │
           ▼                          ▼
┌──────────────────────────────────────────────────────────┐
│                   PROCESSING PIPELINE                     │
│                                                           │
│  ┌──────────┐   ┌────────────┐   ┌─────────┐            │
│  │  FORMAT   │──▶│ VECTORIZE  │──▶│   KNN   │            │
│  │  (clean)  │   │ (TF-IDF)   │   │ (k=5)   │            │
│  └──────────┘   └────────────┘   └────┬────┘            │
│                                       │                   │
│                              ┌────────┴────────┐         │
│                              │  Confidence?     │         │
│                              │  HIGH → classify │         │
│                              │  LOW  → doubt    │         │
│                              └────────┬────────┘         │
│                                       │ (doubt)           │
│                              ┌────────▼────────┐         │
│                              │      LLM        │         │
│                              │  Claude/GPT     │         │
│                              │  "Explain this"  │         │
│                              └────────┬────────┘         │
│                                       │                   │
│                              ┌────────▼────────┐         │
│                              │     LEARN       │         │
│                              │  Active Loop    │         │
│                              │  Retrain model  │         │
│                              └─────────────────┘         │
└──────────────────────────────────────────────────────────┘
           │
           ▼
┌──────────────────────────────────────────────────────────┐
│  📊 OUTPUT: predictions.jsonl                             │
│  {timestamp, line, class, source, confidence, explain}    │
└──────────────────────────────────────────────────────────┘

Philosophy

The alchemists didn't just want to turn lead into gold.
They wanted to understand transformation itself.

Alchemind is the same idea applied to intelligence:
not building smarter machines, but understanding
what intelligence IS — from a mathematical neuron
firing in 1943 to an autonomous system that learns
from every attack it stops.

I started with robots in 2012.
I taught machines to see fire, detect pests, solve CAPTCHAs.
Then I broke them with adversarial attacks.
Then I built systems that can't be broken.

Every chapter of this book is a chapter of that journey.
Every lab is a skill I use in production today.

This isn't theory — this is 13 years of building things
that actually work, compressed into something you can
learn in weeks.

From Patagonia, at the end of the world,
where penguins outnumber developers
and the wind teaches you persistence.

                                    — Daniel Dieser
                                      Puerto Madryn, 2026

Quick Start

# Clone
git clone https://github.com/daletoniris/alchemind.git
cd alchemind

# Setup lab environment
cd labs
docker compose up -d  # PostgreSQL + Jupyter

# Or manual setup
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

# Start with Lab 1
cd 01-classification
jupyter notebook

Requirements

  • Python 3.10+
  • Docker (recommended for labs)
  • No GPU required — all labs run on CPU

About the Author

Daniel Dieser (MrMoz) — Puerto Madryn, Patagonia, Argentina.

Self-taught researcher in AI, cybersecurity, and complex systems. Born in Villaguay, Entre Ríos (1982). Lead Incident Response at a major telco. Designer of hybrid AI systems that combine ML models, LLMs, and human analysts into autonomous defense platforms.

  • Founded AI Resilience Hub Village at Ekoparty — the largest security conference in Latin America
  • Professor at Hackademy (Ekoparty's education platform)
  • Author of "La Realidad que Aprende" manifesto
  • Creator of TokioAI — autonomous SOC platform
  • Published WAF + AI Paper — hybrid WAF with autonomous dynamic learning

The Timeline

2012-2014  Robotics — Building hardware, first experiments
2015-2017  JarvisIA — First Spanish voice assistant
2017       "La Profecía de Villaguay" — newspaper interview
2018       Neural Networks paper — educational publication
2018-2020  Computer Vision — fire detection, NDVI drones, pest detection
2021-2022  Adversarial ML — breaking and defending neural networks
2022       Chubut Hack — co-founded regional security community
2023       AI Resilience Hub — Ekoparty AI Village
2024       WAF + AI Paper — published academic research
2025-2026  TokioAI Platform — production autonomous SOC
2026       Alchemind — this book

Related Projects

Project Description
TokioAI Autonomous SOC platform with 63 tools, multi-LLM agent, Telegram bot
WAF + AI Paper Published paper on hybrid WAF with autonomous learning
adversarial-waf-ml TensorFlow adversarial attacks on WAFs
quantic-encoder Hybrid WAF log classifier with KNN + LLM
agent-smiths MCP-based autonomous security agent
ai-security-ctf AI/ML security CTF challenges
ekoparty-ml-security ML for threat detection training materials

License

This work is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0.

You are free to share and adapt this material for non-commercial purposes, with attribution.


Built from Patagonia 🐧 with persistence, mate, and wind.

tokioia.com · airesiliencehub.com · @daletoniris

About

From the first artificial neurons to autonomous defense systems — Book + Labs + Papers on AI history, neural networks, adversarial ML, and The Warden architecture. By Daniel Dieser.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages