Project Meridian.
Distributed research swarm. Autonomous paper discovery, hypothesis generation, and knowledge graph population across a federated network.
What it does
Meridian sweeps arxiv, semantic scholar, and domain-specific sources continuously. Papers are scored against active research directions, filtered through a woo detector, and ingested into a living knowledge graph. Runs 24/7 across a federated network of machines via NATS messaging.
The pipeline
Continuous sweep → Woo filter (reject pseudoscience) → Relevance scoring against active research topics → Knowledge graph ingestion → Hypothesis generation for human researchers → Zotero library management.
Capabilities.
| capability | what it does |
|---|---|
| Distributed operation | Federated across multiple machines via NATS — runs where the compute is |
| Woo filter | Shared with Agni — pseudoscience detection before it wastes researcher time |
| Topic tracking | Configurable YAML research topics — the swarm knows what you care about |
| Knowledge graph | HippoRAG-powered — findings connect to existing knowledge automatically |
| Zotero integration | Papers scored and catalogued for the research team |
| Cache-to-cache transfer | Knowledge packs built from discovered research will distribute across the fleet via Pharos injection — a discovery-to-deployment pipeline currently in integration testing |
| Daemon mode | Runs every 4 hours unattended — the research never sleeps |
Why it matters
The volume of AI research published daily exceeds any team's reading capacity. Meridian ensures nothing relevant gets missed. The paper that changes your direction shouldn't be the one you happened to read — it should be the one the swarm flagged at 3 AM because it matched a pattern you asked about six weeks ago.
This research is ongoing. If your work intersects with ours —