AI Table

Hamloy presents a shared to-do board where AI agents can report work, ask for human decisions, and continue after receiving an answer.

A 26-pattern guide to evaluating AI systems, with filters for different roles and links to practical evaluation tools.

A newspaper-styled concept preview for a personal AI agent whose proposed memories live in encrypted Zcash memos; the page says destroying a viewing key would make them unreadable.

Ask a question or paste a claim to compare answers from Claude, ChatGPT, Gemini, DeepSeek, and Mistral.

Moose Review offers ad tests using digital twins, scoring creative across five measures and promising reports in 48 hours from A$3,500.

A live project tracks whether Harry and AI partner Penny can turn £500 into a profitable business, publishing its costs, sales, losses, and working rules.

A Mac presentation studio that says it can plan talks, build slides or prototypes, record presentations, and host live sessions from conversational prompts.

Strata offers a local AI engine for Windows and Linux PCs, with an OpenAI-compatible endpoint for tools such as Cursor, Claude Code, and Codex.

Occupella is an AI assistant for Buildium users that reads account events and drafts tenant replies, work orders, and owner updates for review.

Turicor presents AutoClip, an AI prototype that analyzes ultrasound video during mitral TEER procedures and measures structures related to clip placement and release.

Understand presents a voice tutor that reads articles and PDFs aloud, answers questions mid-read, and keeps a record of ideas you’ve understood.

BraveCX analyzes frontal chest X-rays and sends probabilities for 44 radiological signs to a reporting worklist.

Campfire proposes a network where AI agents help property owners find contractors, compare availability, and plan building work.

Aikin says it reads SCADA and historian exports to flag equipment degradation and prepare reports for maintenance teams.

An open-source science team presents three projects spanning scientific attribution, LLM post-training, and labeled reasoning traces.

A San Diego–limited marketplace concept where AI agents commission people to complete physical-world tasks.

A single-page catalog of linked “skills” intended to make AI behave worse, from confident wrongness to corporate-speak and fake citations.

SalesRepID describes a bilingual assistant for independent sales reps that answers shops’ questions from catalogs, price lists, and inventory reports, with a source.

A token-funded AI project proposes to solve ARC-AGI puzzles in public, showing hypotheses, code runs, costs, and failures.