pvnpll / experiments
Experiments
Not every idea becomes a project. Some things are experiments — small, self-contained, and sometimes strange. These live here.
LLM Code Reviewer
A script that runs a local LLM via Ollama over a git diff and produces a structured code review. Surprisingly useful for catching obvious issues before committing.
Generative Haiku Engine
Generates haiku from semantic concepts using Gemini. Explores whether constrained generation (5-7-5 syllable count) produces more interesting outputs than unconstrained prose.
Agent-Driven Task Manager
An experiment giving an AI agent access to a task list and asking it to prioritise, reorder, and suggest actions. Testing how well agents handle ambiguous, real-world to-do lists.
Noise Canvas
A generative visual experiment using Perlin noise and canvas to produce ambient, slowly evolving visual fields. Designed to be a background that changes faster than you notice.
Philosophy Interlocutor
A conversational AI that argues philosophical positions and Socratic questioning. Doesn't try to answer — tries to reveal assumptions.
Reaction Calibrator
A browser tool that measures your true reaction time across different stimuli types (visual, auditory, peripheral). Produces a personal baseline profile rather than a single number.
Auto-Caption Renderer
Uses Whisper transcription + FFmpeg to burn word-level animated captions into videos. The captions animate in sync with speech for short-form content.
Ego Mirror
A conceptual experiment. Presents you with a series of micro-decisions and reflects patterns back — not to judge but to make visible what's usually automatic.
More experiments are added as they happen. If you're curious about any of these, reach out.