A personal portfolio website that doubles as a control and monitoring hub for my automations:

• Server monitoring - live and historical hardware data readings from the Raspberry Pi server, charted over a selectable time range.
• Video automation - a dashboard covering each generation run and the performance of published videos.
• Home control - a dashboard for configuring smart devices based on the live electrical output of the solar panels.
• Registration - a dashboard for selecting preferred booking times for external services, which are then reserved automatically within minutes of registration opening.
• Access control - role-based access control, with invite-only registration.

Generates and publishes short-form videos from a per-account configuration that defines the theme, tone and narration style. It runs daily with no human involvement, and every external service it relies on sits on a free tier:

• Script - Gemini 2.5 Flash writes the script from the account's theme, returning the narration, title, description and tags as JSON.
• Voice - the script is voiced by one of three text-to-speech providers: Cartesia, ElevenLabs and Unreal Speech, arranged in a fallback chain. When a provider repeatedly fails or runs out of credits, the run moves on to the next set of keys, and then to the next provider.
• Footage - a background clip is picked from a pre-selected local library, and a random section of it is trimmed to the length of the audio.
• Subtitles - Vosk transcribes the audio into word-level timings, running locally.
• Render - FFmpeg burns the captions into the video, adjusts the resolution and pacing for vertical playback, and splits it into shorter parts, only ever cutting where a sentence ends.
• Publish - the videos are uploaded through direct API calls and automated Selenium scripts. Browser profiles persist between runs, so sessions stay authenticated and no daily logins are needed.
• Tracking - each account is tracked on its own and reports to a PostgreSQL database.
• Error handling - a fatal error stops only the account it occurred in rather than cascading through the rest of the run, and authenticated users can review the full error detail on this site.

The whole thing runs on a Raspberry Pi 5. You can follow it live on the status page: Video Automation shows how each run went, and Video Statistics tracks how the published videos are performing.

Automation that restores Discord's voice and notification sound levels in the Windows volume mixer, which reset on every launch due to multiple audio processes. A one-time install sets up a scheduled task that runs whenever Discord starts, identifies each process, and applies the predefined sound levels.

Visualises different pathfinding algorithms on interactive user-drawn mazes.