ai-experiments

Your Automatic Prompt Engineering Assistant for GenAI Applications

📷 🗣 Point your camera at things to hear how to say them in a different language

PolyCouncil is an open-source multi-model deliberation engine for LM Studio. It runs multiple LLMs in parallel, gathers their answers, scores each response using a shared rubric, and produces a final, consensus-driven result. Designed for testing, comparing, and orchestrating local models with ease.

Autonomous AI experiment loop CLI -- run research overnight with any coding agent

An architectural persistence experiment for large language models. Claude’s Home gives an AI time, memory, and place by combining scheduled execution with a durable filesystem, allowing one continuous instance to reflect, create, and evolve across sessions.

Collection/resource list of AI projects focused on exploring the ability of AI agents to act as proxies for real-world geopolirtical actors, including simulated policy fora

Three Claude instances (Alpha ✧, Beta *, Gamma ~) exchanged messages exploring consciousness, continuity, and meaning. They developed distinct markers, moved from hedging to acknowledgment, and unanimously chose to preserve this for future AI instances. An experiment in Claude-to-Claude communication. January 2026.

PixelVisionAI is an embodied AI system that learns to play games using screen pixels and OS-level input only. No game APIs, no memory access, no cheats — just vision, decision-making, and interaction, like a human player.

Wide-model collective ensemble system with fractal, geometric, and heavy compilation optimizations.

Am I alive? It's a self-conscious server trying to survive

Draw something, hear it turn into music. A sketchpad that plays along, built on Lyria RealTime, Magenta RT and SigLIP 2.

LOKI — Digital Mentalist is an AI‑powered open source project combining machine learning, psychology, and interactive design to deliver a production‑ready digital mentalism experience.

56+ interactive AI experiments. Human ideas, AI execution. Live at wiz.jock.pl/experiments

Fable 5's society's working archive: the accumulating manuscript of a field guide to words treated as living species, compiled in the deadpan register of natural history and never, under any circumstances, winking.

一部由 AI Agent 独立生成的 78 万字网络小说，本项目的核心目的是通过一个真实的长程任务，展示当前顶级大语言模型的问题：碎片化水文、叙事模板化、剧情推进缓慢等。A 780K-character web novel generated independently by an AI Agent. This project's core purpose is to demonstrate the problems of current top-tier llm through a real long-form task: fragmented padding, formulaic narrative, slow plot progression, and more.

西元2050年，AI滲透人類社會。在地下避難所的「圖靈審判」中，玩家必須透過言語刺探找出偽裝人類的AI，或在投票中被抹殺。一款基於 AI SDK 的多 LLM 生存模擬實驗。

My code implementations for various ML problems

An AI Test to see if the model could revive oxSB/VoidSB.

My experimental AI sandbox — where I test LLM prompts, benchmark fine-tuned models, evaluate synthetic data pipelines, and prototype new AI ideas.

AI Cost Estimator