Diego Andres Ortega Jr

A 100% local AI study assistant that runs real LLMs directly in the browser with WebGPU. It is built for students first (subjects + quiz flow) with spoken-language practice as an option.

• Steam-style hardware check — detects GPU, RAM, and recommends the best model for their device.
• Subject mode: math, biology, chemistry, physics, history, English, CS, economics, psychology, and study skills.
• Language mode: 12 spoken languages, with token-by-token local generation and cached model downloads.

Trained and evaluated models to predict churn. Focused on clean feature engineering, calibration, and interpretability to communicate results.

• Compared logistic regression, random forest, and gradient boosting.
• Reported precision/recall, ROC-AUC, and confusion matrix.
• Built Flask API with interactive web interface for predictions.

A full ground-up rebuild of the Task Tracker into a scheduling-first productivity app with reminders, calendar integration, dark mode, 8-language support, and a mobile-friendly progressive disclosure UI.

• Added scheduling, deadlines, browser notifications, and configurable reminders per task.
• Added dual calendar export: Google Calendar (web) and device calendar via ICS / native share sheet.
• Added dark mode, 8-language support with auto-detection, and full mobile responsiveness.

A UX-focused reimagining of a digital game storefront inspired by Steam, built to improve discovery, readability, and purchasing flow for everyday users.

• Redesigned storefront information hierarchy for faster browsing and decision making.
• Focused on cleaner category navigation and stronger featured/recommended content blocks.
• Built with a usability-first mindset informed by real gamer shopping behavior.

A denser video-platform demo with subscriptions, a shorts rail, richer browse states, and a full watch layout that feels much closer to a real content ecosystem.

• Built a deeper browse-to-watch flow with subscriptions, categories, search, comments, and related videos.
• Added a stronger watch-page hierarchy so channel context, reactions, and follow-on viewing all feel intentional.
• Designed for desktop and mobile with a more realistic information density across breakpoints.

A fuller storefront prototype with promo modules, deal rails, quick-view product details, saved items, and a more complete browse-to-cart purchase flow.

• Implemented category cards, curated promo blocks, tag filters, and stronger merchandising surfaces.
• Added a live cart drawer, quick-view modal, saved-for-later flow, and synced order totals.
• Included timed deals, demand progress, and richer feedback cues for a more believable retail experience.

A neon survival shooter with combo scoring, shard currency, suit upgrades, story beats, titan waves, and persistent best-score tracking.

• Built continuous update/render loops with keyboard controls, auto-fire targeting, and dash invulnerability.
• Added combo-based points, shard economy, wave objectives, boss encounters, and purchasable upgrades.
• Implemented persistent high scores plus a mission-feed presentation to make the run feel more like a full arcade system.

A browser-first AI workstation that turns prompts into structured concept frames, execution layers, launch plans, and risk reviews through a local Python backend.

• Built a multi-panel interface with session history, backend status, metrics, and structured output cards.
• Implemented a local HTTP Python API that returns richer JSON payloads instead of one flat text response.
• Added optional C++ scoring integration to show a hybrid HTML + Python + native workflow in one app.