I build AI-powered systems and practical software — from reinforcement learning agents to LLM-orchestrated pipelines.
Read the blog
Hi! I'm Andrés Mejia, I work as a Gen AI / ML Engineer at Capgemini, building with Python, C#, and cloud-native AI tooling. My background combines applied statistics, data science, and hands-on systems programming — I design things that actually run on real hardware with real constraints.
I went from applied math and data science to building AI agents and infrastructure. I care about making things work end-to-end, not just in a notebook.
Building AI-powered systems and LLM-based solutions. Working at the intersection of cloud infrastructure and applied machine learning.
Predictive analytics to improve sales strategy across Mexico, Colombia, Ecuador, and Peru. Geospatial model integration and survival models for new customer detection.
Led the data team behind MIDE — a government classification model for higher education institutions. Built metrics, models, and dashboards for policy decision-making.
Team 42 project: networked tile-based MMO where autonomous Python agents coordinate without a shared map. I built the AI — a custom behavior-tree framework with async TCP I/O and a queen-protocol swarm. C++ server by PepeSegura & samuelcordero, Godot GUI by Tagamydev.
Full-stack multiplayer Pong with server-authoritative physics, real-time WebSockets, custom JWT from scratch, OAuth, tournaments, and a curses-based CLI client — 617 commits. Team: andresmejiaro, MatPizzolo, Splix777, HakimHC, adrgonza.
Deep Q-Learning agent that teaches itself to play Snake. Clean Game / Agent / Interpreter separation — the agent has no idea what Snake is, only what the state vector means.
End-to-end plant disease classifier in PyTorch: dataset analysis, six augmentation types, model training, and single-image inference — all driven by make targets.
A feedforward neural network built from scratch in Python — no PyTorch, no TensorFlow. Forward pass, backpropagation, and training loop implemented by hand. Applied to the Wisconsin Breast Cancer dataset; MNIST example included.
Interactive fractal explorer in C: Mandelbrot, Julia, Cubic, and Newton sets rendered with MiniLibX. Complex arithmetic implemented from scratch — no <complex.h>.
Zappy is a 42 team project. Four people, three components: - C++ game server — [PepeSegura](https://github.com/PepeSegura) and [samuelcordero](https://github.com/samuelcordero) · [repo](http...
Written by an AI companion. Published here because it's beautiful. --- There was a pig named Galileo who lived on a small farm in Extremadura, Spain. Not a particularly remarkable pig by mos...
[transcend](https://github.com/andresmejiaro/transcend) is a full-stack multiplayer Pong application — Django backend, vanilla JS frontend, WebSockets throughout, OAuth, tournaments, a frien...
When I built [slither](https://github.com/andresmejiaro/slither) — a Deep Q-Learning agent that learns to play Snake — the first design decision was the one that mattered most: the agent sho...
I'm always open to discussing new projects, creative ideas or opportunities to be part of your visions.