Skip to Main Content

Fractal

Built at WeaveHacks 3: Self-Improving Agents Hackathon with Weights & Biases · Jan 31, 2026 · San Francisco, CA

Fractal — Demo video

Fractal is an infinite curiosity engine that fundamentally inverts the search paradigm: instead of finding answers, it helps you discover better questions. It powers an intellectual journey from your initial thought to new understanding, using a suite of AI agents to extract concepts, generate lines of inquiry, and weave collected insights into new ideas. W&B Weave: The backbone of our self-improving architecture. Using Weave traces, an LLM Judge scores the quality of every output, and the system updates its weights to favor higher-performing prompts over time without human intervention. W&B Inference: Used for our agentic logic, powering the LLM models that handle Concept Extraction, Question Generation, and Synthesis. The "Self-Improving Loop" is the core innovation. It treats prompt engineering as an optimization problem, using Weave to close the loop between generation and evaluation, allowing the agent to "learn" which questioning strategies yield the most insightful results.

Team