Skip to content
View dgsolera-ai's full-sized avatar
  • Alicante, Spain

Block or report dgsolera-ai

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
dgsolera-ai/README.md

Hey, I'm Dennis

Final-year AI Engineering student at the University of Alicante.

I'm interested in what happens after the notebook: how models get deployed, monitored, and what they actually do to the people on the other side of their decisions.

Currently:

  • 🔬 Working on fairness auditing and explainability for ML systems
  • 🏗️ Building end-to-end data pipelines (Airflow · Spark · Delta Lake)
  • 📚 Preparing my final year thesis
  • 🌱 Learning MLOps in depth — FastAPI, Docker, model monitoring

🔧 Stack

Languages — Python · SQL · Bash
ML — PyTorch · scikit-learn · LightGBM · SHAP · Fairlearn
Data — Spark · Airflow · Delta Lake · MinIO · pandas
Infra — Docker · Linux · Git


📌 Featured

hmda-fairness-audit — Detecting and mitigating racial bias in mortgage lending decisions across 283K real loan applications. Cut the racial approval gap from 34.6 to 13.7 percentage points using Fairlearn, and quantified exactly what it cost in model performance.

lunarlander-policy-gradients — Policy gradients on LunarLander. An entropy bonus set 10x too high taught the agent to hover forever instead of landing — reward looked fine, episode duration gave it away. Diagnosed and fixed.

graph-neural-networks — From GNN fundamentals to an NP-hard problem: a controlled study of when graph structure actually helps (GCN goes from 0.53 to perfect accuracy depending on homophily), then solving Maximum Clique with a GNN + Replicator Dynamics hybrid.

llm-systems-from-scratch — The modern LLM stack built from scratch on a 1.5B CPU model: prompt engineering, RAG (with hybrid semantic + lexical search), and LangChain agents. Focused on where each technique breaks, not just where it works.

vae-wae-generative-models — Generative models from scratch: VAE (KL) vs Wasserstein Autoencoder (MMD), compared across 2D images and 3D voxel grids (MNIST, CIFAR-10, ModelNet10). Focused on where each latent regularizer breaks, not just where it works.


📫 Reach me

LinkedIn · dennisgarciasolera@gmail.com

Open to internship opportunities — ML in production, data engineering.

Pinned Loading

  1. lunarlander-policy-gradients lunarlander-policy-gradients Public

    Policy gradient methods on LunarLander-v3: REINFORCE, Actor-Critic, and what happens when your entropy bonus teaches the agent to hover instead of land.

    Jupyter Notebook

  2. hmda-fairness-audit hmda-fairness-audit Public

    Forked from jcm204/aa-proyecto-hmda-ny

    Detecting, quantifying and mitigating racial bias in mortgage lending decisions (HMDA New York 2024) with LightGBM, SHAP and Fairlearn.

    Jupyter Notebook

  3. graph-neural-networks graph-neural-networks Public

    From GNN fundamentals to solving an NP-hard problem: message passing, over-smoothing, and Maximum Clique with a GNN + Replicator Dynamics hybrid

    Jupyter Notebook

  4. llm-systems-from-scratch llm-systems-from-scratch Public

    A hands-on progression through the modern LLM stack on a 1.5B CPU model: prompt engineering, RAG from scratch, and tool-using agents. No APIs, no GPU.

    Jupyter Notebook

  5. vae-wae-generative-models vae-wae-generative-models Public

    VAE (KL) vs Wasserstein Autoencoder (MMD) compared across 2D images and 3D voxel grids — MNIST, FashionMNIST, CIFAR-10, 3D-MNIST, ModelNet10. Focused on where each regularizer breaks.

    Jupyter Notebook