I design, deploy, and operate production-style AWS infrastructure — serverless architectures, secure IAM boundaries, and the systems that run them. This portfolio is itself one of those systems: everything below is live infrastructure, not a mockup.
I build the cloud systems that power production applications — and this site runs on one of them.
Every project below runs on its own subdomain, its own AWS resources, and its own scoped IAM role — isolated by design, not by convention.
Privately-hosted static site behind CloudFront, with Origin Access Control, a Lambda-driven contact form via SES, and a documented decision log for every architectural choice.
static-site.luigisantoro.dev →A task manager backed by API Gateway, Lambda, and DynamoDB, with Cognito-authenticated users — each user's tasks isolated by partition key.
api.luigisantoro.dev →Monitors uptime across this portfolio on a schedule — Step Functions orchestration, parallel checks, and SQS-backed alerting when something goes down.
status.luigisantoro.dev →Four polyglot microservices on Kubernetes, communicating through a choreographed Saga over SQS. Infrastructure is on-demand — not permanently live, given EKS's fixed hourly cost. Full architecture, decision log, and automation scripts are in the repository.
github.com/luigisantorodev/ecommerce-platform →Predicting appliance energy usage from indoor/outdoor sensor data. Four models compared, from linear regression to Random Forest, with full diagnostics and honest reporting of what didn't improve results.
ai.luigisantoro.dev/energy-regression →Classification under extreme class imbalance (0.17% fraud rate). SMOTE and class weighting compared across two model families, threshold tuning, and case-by-case explainability with SHAP.
ai.luigisantoro.dev/fraud-detection-classification →Unsupervised customer segmentation via RFM analysis. K-Means clustering cross-validated with hierarchical clustering, three segments visualized with PCA.
ai.luigisantoro.dev/customer-segmentation-rfm →Unsupervised intrusion detection on NSL-KDD, no labels used in training. Isolation Forest vs One-Class SVM, tested against never-before-seen attack types.
ai.luigisantoro.dev/network-intrusion-anomaly-detection →Time series forecasting compared across ARIMA, Prophet, and Random Forest — including why two structurally different models failed in the exact same way.
ai.luigisantoro.dev/nasdaq100-forecasting →A CNN scored a perfect 100% on visual defect inspection — so I audited the test set for data leakage instead of taking the number at face value.
ai.luigisantoro.dev/industrial-defect-detection-cnn →Transfer learning with a frozen ResNet18 — training only 0.046% of the network to classify satellite imagery into 10 land use categories.
ai.luigisantoro.dev/satellite-landuse-transfer-learning →Comparing Whisper model sizes on clean vs. challenging audio — Word Error Rate weighed against inference time, not accuracy alone.
ai.luigisantoro.dev/whisper-speech-to-text-comparison →Transfer learning on Mel spectrograms to classify emotion in speech — plus a reproducibility bug caught and fixed before it could bias the result.
ai.luigisantoro.dev/speech-emotion-recognition →A hand-built Canny/Hough pipeline against a transfer-learned U-Net — including a diagnosed structural failure on Botts' Dots reflectors and a reworked evaluation metric to compare a line against a mask fairly.
ai.luigisantoro.dev/lane-detection-cv-vs-unet →A consistent set of principles across every project here, regardless of stack.
Every project runs under a dedicated role with only the permissions it needs, assumed via AssumeRole — never a broad, standing credential.
Each project has its own subdomain and its own AWS resources. A failure or compromise in one project doesn't extend to another.
Architectural choices are recorded where they're made, with the reasoning behind them — including the ones that changed along the way.
This site and the projects it showcases were engineered alongside Claude, reflecting the same AI-integrated workflow I rely on in professional engineering practice.
Send a message and it comes straight to me. If you'd like, you can also receive a technical breakdown of how the Serverless Static Site project is built.
