Skip to content
MetademicMETADEMICRESEARCH LAB
Roadmap

Concept → Sustainable Service

RACoN from concept to WAN — each validation stage maps to a research publication. Hybrid P2P: prompt → coordinator picks best GPU → peer generate().

How RACoN works →Research index
Concept→Simulation→Prototype→LAN→Distributed AI→Security→WAN→Scale→Public Preview→Global Hybrid→Sustainable Service
3
Phases
2022 — 2026
7
Publications
Papers 1–7
P2P
Hybrid mesh
coordinator + peers
2022FOUNDATION Completed
1 publication
Development / Validation
  • Concept + prior-art validation
  • Research gap + reference architecture
  • RACoN-Sim design begins
Milestone: concept scientifically defensible + simulator specification
#01
PAPER 1 — Architecture & Prior Art
Maps existing systems and defines RACoN's research gap, reference architecture, feasibility boundaries, and core validation questions.
Forthcoming preprint
2025BUILD & VALIDATE Completed
3 publications
Development / Validation
  • Complete RACoN-Sim
  • Python RACoN Alpha
  • LAN hardware sharing
  • Distributed LLM development
Milestone: real jobs shared across heterogeneous lab computers
#02
PAPER 2 — RACoN-Sim
Introduces a reproducible simulator and compares scheduling policies under heterogeneous compute, latency, churn, reliability, and demand.
Forthcoming preprint
#03
PAPER 3 — RACoN Alpha
Presents the Python peer/coordinator framework and evaluates deployment reproducibility, hardware-backend compatibility, job dispatch, and system overhead.
Forthcoming preprint
#04
PAPER 4 — LAN Agent Scheduling
Evaluates capability-aware agent-task distribution across real lab computers, measuring speedup, utilization, queueing delay, reliability, and energy.
Forthcoming preprint
2026BUILD & VALIDATE In progress
3 publications
Development / Validation
  • Distributed LLM validation
  • Security + failover + credits
  • Trusted WAN deployment
  • Large-scale test environment
Milestone: trusted multi-location RACoN survives faults and WAN conditions
#05
PAPER 5 — Distributed LLM Inference
Tests topology-aware model partitioning across heterogeneous LAN devices and identifies when distributed inference outperforms single-node baselines.
Forthcoming preprint
#06
PAPER 6 — Trust, Recovery & Credits
Evaluates checkpointing, failover, sandboxing, adaptive verification, and reciprocal credits under node failures and malicious behavior.
Forthcoming preprint
#07
PAPER 7 — RACoN-WAN
Studies real Internet deployment across locations and ISPs, quantifying latency, NAT, packet loss, churn, relay overhead, and regional scheduling effects.
Forthcoming preprint

Papers 1–7 map directly to the validation stages above. See Research index for forthcoming preprints.

How RACoN worksOpen Research