AI / ML · Prototype

Multi-Agent Research Engine

Planner, searcher, critic and writer producing cited reports.

In plain words. Ask a big question and a team of AI roles plans the research, searches the web, checks the sources and writes a report with references.

The problem

One search and one completion rarely answers a multi-part question, and fluent text hides which claims have a source.

What it does

A four-stage workflow: the Planner splits the question, the Searcher gathers sources, the Critic checks coverage and can request more searching, and the Writer produces a report with inline references. Progress streams to the UI.

Key decisions

Responsibilities are split into stages so each intermediate artifact is inspectable.
The critique loop is bounded to avoid endless searching.

Limits and status

Critic judgements are model output and need external evaluation.
No accuracy or time-saving claim is made.