A portfolio of analytical applications spanning financial markets and global news — built with Python, artificial intelligence, and real-world data.
By senior AI engineer Fernando Borbón, PhD
Free hands-on Jupyter courses in English and Spanish. Neural Networks 1001 goes from a single neuron to a GPT on a laptop CPU, then a capstone on real defect-inspection photos. Python 1001 runs from data structures and PySpark to testing, forecasting and a Kaggle capstone in 13 notebooks.
Apartments that breathe tranquility in beautiful Costa Rica. Browse the apartments, amenities and location.
Spain OMIE day-ahead price forecasts, a walk-forward backtest, and a battery-arbitrage schedule from a linear program.
Technical write-ups of real, deployed AI systems: architecture, real data, what it cost to run, and what broke.
Today's global news digest on an interactive world map, with breaking news, regional deep dives, a Social Pulse tab with AI search and an AI News tab: top AI blogs and the month's most-cited papers.
Correlation-based BUY / SELL / HOLD signals for the full S&P 500, with 7-day price targets and fundamentals.
LangGraph agent forecasting real UK wind farm output and diagnosing turbine efficiency against the Betz limit.
No services match your search.
Free hands-on Jupyter courses in English and Spanish. Neural Networks 1001 goes from a single neuron to a GPT on a laptop CPU, then a capstone on real defect-inspection photos. Python 1001 runs from data structures and PySpark to testing, forecasting and a Kaggle capstone in 13 notebooks.
Apartments that breathe tranquility in beautiful Costa Rica.
Technical write-ups of real, deployed AI systems — architecture, real data, what it cost to run, and what broke along the way. Not generic AI commentary.
Multi-agent AI pipeline: 22 countries, 66 RSS sources daily, summarised by Amazon Nova on AWS Bedrock into a news site with a world map, breaking news, regional deep dives, a Social Pulse tab with RAG search and an AI News tab (top AI blogs, monthly most-cited papers). ~$1/month.
Correlation-based trading signals for the full S&P 500 universe. Predicts 7-day price targets using Random Forest regression on daily returns, with interactive charts and fundamental analysis.
Personal cloud backup tool syncing an external hard drive to AWS S3 Glacier Instant Retrieval. Incremental sync with SHA-256 and move detection (no re-upload), plus a private Explorer-style web UI to browse, preview, and download any file via time-limited pre-signed URLs. ~$10/month.
Multi-tenant job search platform unifying discovery, CV building, and application autofill. An autonomous multi-agent system searches 33 connectors daily, ranks listings via embeddings and an LLM agent, then AI-tailors a CV and letter injected via Job Filler. Free through Gold via Paddle.
Private cross-device Kanban board with drag-and-drop task management, custom categories, urgency levels, file attachments (PDF, CSV, images), location and datetime fields, and a month-view calendar. State synced across devices via DynamoDB. Fully serverless: S3 + CloudFront + Lambda. ~$0.41/month.
Consolidates exports from multiple bank accounts into one interactive view. Auto-categorises transactions with keyword rules, persists manual overrides, and uses linear regression to predict next-month income and expenses per category. Runs entirely locally, no cloud services, no APIs.
Personal family photo viewer exploring 194 000+ photos across a world map and timeline. Private tagging, favorites synced across phone and web, phone-to-cloud album sync, and slot-machine discovery. Fully serverless: S3 Glacier IR, CloudFront, Lambda EXIF, React PWA, Android app. ~$2-3/month.
Traffic dashboard for this site, its demos, and my GitHub repositories — daily visitors per site, visitor countries, bot traffic share, and repo views/clones, built from nginx access logs and the GitHub traffic API.
Self-hosted WireGuard VPN on AWS EC2 with AdGuard DNS ad blocking. PyQt6 desktop tray app, mobile web PWA, and a fully standalone Android APK that connects directly to your own server — no desktop, no Wi-Fi dependency. ~$0/month incremental cost.
Real BTC/USD price and on-chain fundamentals, a real bounded window of whale transactions scanned live from Bitcoin blocks, a generalized ESD anomaly detector, and an honest lead-lag correlation test between whale activity and price returns.
Detects facial expressions from YouTube videos using DeepFace for fast emotion classification and Claude Vision AI to explain why each face shows that specific feeling. Jupyter notebook demo.
Real Spain OMIE day-ahead prices, a walk-forward backtested forecasting model, two honest negative feature-engineering results, and a battery-arbitrage optimizer (linear program, PuLP/CBC) — against a reference battery, since no real one exists yet.
LiDAR point clouds, Sentinel-2 imagery, and OpenStreetMap ground truth fused into a RandomForest building extraction pipeline over Washington DC, evaluated with pixel and polygon-level GIS metrics (66% IoU), plus a Bedrock Nova benchmark testing an AI's grasp of GIS and remote sensing concepts.
Downloads songs from a public-domain YouTube playlist and plots energy and stress curves per track using librosa. Overlays all songs to compare the most intense vs calmest pieces side-by-side. Jupyter notebook demo.
Interactive breakdown of Spain's consolidated public spending across all government levels — sourced from IGAE and Eurostat COFOG data. Explore exactly where your taxes go by category.
Raspberry Pi noise station: live dB(A) map, clips of nights over the limit with a live spectrum, and an Add device form that builds the SD card and can locate it by WiFi. Demo: five simulated stations, from New York traffic to Amazon parrots and a Madrid goal.
Deep Q-Network agent trained to play Pac-Man. Tackles a state space of 2²⁴⁴ dot configurations using function approximation, delayed credit assignment, and context-dependent reward shaping. Watch the AI navigate ghost evasion in real time.
LangGraph agent forecasting real wind farm production (real UK SCADA data, three farms) and diagnosing turbine efficiency against the Betz limit. RAG over a real fault log and over real turbine datasheet and tender PDFs with page citations, self-hosted MLflow, live on AWS.