Gallery
1 / 3
What it does
EINSTEIN Matchmaker is a multi-agent system on Mozaik that matches company challenges from the EINSTEIN network with university researchers and writes a concrete collaboration proposal for each pair. Jev screens every researcher for a challenge, a Matchmaker picks the pair, a Writer drafts the proposal and a Critic checks it against the sources and sends it back until it holds. Every fact is quoted from the data and checked by code, and weak pairs are dropped with a reason instead of being forced.
How the agents run concurrently
Several company challenges are processed at the same time, each in its own workflow on a shared Mozaik runtime. Within a challenge, screening is split into batches of researchers, and each batch is a separate Jev agent; all batches run in parallel. Inside the Critic, the LLM review agent and the Jev fact-check agent work on the same draft at the same time, and code combines their findings. The pairs of one challenge are tried one after another, because the next pair is needed only if the current one is dropped. LLM requests share one runtime that limits how many run at once and stops at a budget, so many workflows can run without exceeding the spend.