Multi-Agent Swarm
Watch a planner decompose a goal, parallel worker agents (across models) execute, and a judge synthesize — live orchestration graph.
Live demos
Original systems that actually run.
Each is built from scratch — not client work — and opens its own page with a live, visual demonstration of what the system is doing. They call live models server-side and are rate-limited.
Deep-Research Agent
Plans sub-questions, gathers evidence with tools, and writes a cited report — with a live research trail + sources panel.
Reflexive Agent
Drafts an answer with tools, critiques its own draft, then revises — the self-correction loop, streamed.
Synthetic Data Factory
Generate → auto-grade → dedup → filter a dataset, with a live quality dashboard. The loop a real curation pipeline runs.
Auto-Rubric Evaluator
Auto-builds a weighted rubric, runs two models, and grades each per criterion with an animated score matrix.
Multimodal Cross-Checker
Checks whether an image supports a claim and flags misleading charts — with a verdict badge + confidence gauge.
Training-Run Automation
An agent picks hyperparameters, then a real classifier trains with a live loss + accuracy chart and held-out eval.