Things I've built

This website: a Next.js portfolio in three languages with an AI assistant that answers questions about my career in my own voice, grounded on the same typed content that renders the pages.

Challenges

Keeping the assistant accurate (no hallucinated facts), cheap to run and safe to expose publicly, while keeping a single source of truth for both the UI and the model's knowledge.

Results

Rebuilt in 2026 from a LangChain + vector-database RAG pipeline into a leaner design with long-context grounding, prompt caching, tool calling with generative UI, rate limiting and automated evals — fewer moving parts, lower cost and better answers.

• Next.js
• React
• TypeScript
• Vercel AI SDK
• OpenAI
• Tailwind CSS
• next-intl
• Vercel

Built at the Polo Digital de Málaga Data Challenge: a system that cross-references the city's open datasets to find the neighbourhoods under the most pressure for used-cooking-oil recycling — lots of generation, few collection points nearby — and a service concept with containers in building entrances and door-to-door pickup, turning household oil into biodiesel.

Challenges

Finding, cleaning and joining heterogeneous public datasets under hackathon time pressure, and turning them into a clear, actionable map and ranking for communities and recycling companies.

Results

Third prize. A hexagon heat map and neighbourhood ranking of recycling pressure, plus a landing page for the pilot, helping communities decide where to install containers and connecting the oil with companies that give it a second life. I worked on the technical development in a team of five.

• Open data
• Data analysis
• Geospatial visualization
• Smart cities

A Flutter mobile app that helps people cook more efficiently: find and save recipes, build shopping lists, and follow step-by-step instructions hands-free with voice commands.

Challenges

Integrating accurate voice recognition with limited resources and designing an interface that works smoothly across devices.

Results

Hands-free cooking flow powered by voice commands, plus experience in responsive mobile UI design and API integration.

• Flutter
• Dart
• Speech recognition
• REST APIs

A restaurant and reservation management web app with an interactive editor to design restaurant floor plans, real-time chat between customers and restaurants, deployed on AWS.

Challenges

Building an intuitive floor-plan design interface and reliable real-time communication through the chat system.

Results

A user-friendly layout tool, solid experience with real-time features, and hands-on AWS deployment and cloud management in an international team.

• React
• Express.js
• Node.js
• WebSockets
• AWS

A Unity game where an AI agent autonomously pilots a spaceship with physics-based controls, using reinforcement learning to dodge asteroids, destroy threats with guided missiles and chase UFOs.

Challenges

Training the agent to control movement through physical forces, making real-time decisions to dodge obstacles and fire missiles, and keeping spaceship and missile behavior physically realistic.

Results

An autonomous navigation agent that adapts to space hazards, dodging or destroying asteroids and pursuing UFOs. Hands-on experience with reinforcement learning, physics-based control and real-time decision-making.

• Unity
• C#
• Reinforcement Learning
• ML-Agents

A deep-learning model built with TensorFlow and Keras that estimates a person's age from a photo using convolutional neural networks (CNNs).

Challenges

Handling a large dataset of low-resolution images, transforming it into a format suitable for training, and fine-tuning the network for accurate predictions.

Results

A working age-prediction model and a research-style paper. Deeper understanding of CNNs, image processing and model optimization, plus experience training on cloud environments like Colab and AWS.

• Python
• TensorFlow
• Keras
• CNNs
• Google Colab
• AWS