Track 02
The University of Novi Sad has thousands of researchers and NTP Novi Sad hundreds of companies, yet they rarely work together. Build agents that find the most promising researcher–company pairs and write a concrete proposal for each.
The roles below are our suggestion, not a requirement. Split the work however you think is best; what matters is that several agents share it and check each other's work.
| Agent | Role |
|---|---|
| Orchestrator | Takes the task, hands out work, tracks progress, assembles the result. |
| Research Profiler | Builds researcher profiles from public sources (ORCID, OpenAlex, CORDIS, Google Patents, Lens) or the dataset: fields, skills, projects, patents, each with its source. |
| Industry Analyst | Pulls each company's technical challenges, needs and direction. |
| Matchmaker | Ranks pairs and explains each: why these two, and which problem they solve together. |
| Proposal Writer | Drafts an industrial PhD topic or a one-page joint project for each chosen pair. |
| Critic | Checks every claim against its source and sends unsupported ones back. Never accepts invented papers or skills. |
More ideas: a Funding Scout (open calls such as Horizon Europe or the Innovation Fund), an Outreach agent (an intro email for each pair), Reverse Search (a company describes a problem, the system finds researchers), or a step where a person approves pairs before the Writer starts.
A synthetic dataset we made for the hackathon: fictional researchers and companies, with publications, projects, patents and the companies' challenges. A guide to the open APIs, an example of a good industrial PhD topic, and mentors from UNS and NTP throughout.
An architecture diagram, the trace of one full run showing where the Critic stepped in, and at least three proposals for three different pairs. In the README, answer one question: how would this plug into the EINSTEIN network and EINSTEIN Stars?
Data: only public data or the provided dataset. The system proposes collaborations; it never scores or ranks people by worth.
| Criterion | Points |
|---|---|
| Multi-agent architecture: roles, coordination, review loops | 30 |
| Proposals a company and a researcher would actually sit down over | 25 |
| Accuracy and transparency: sources, trace, no hallucinations | 20 |
| Potential for real use in EINSTEIN | 15 |
| Responsible use of data | 10 |