Overview
# EDGE-RAN Gary: AI-Native Radio for Equitable 6G Access
7GC · STAGING_WORKER
AI-RAN Gary
Research repository for an AI-native RAN controller and equitable spectrum allocation experiments.
Interactive research app
Explore AI-RAN Gary
Run the live Streamlit research interface directly here. The application is hosted on Streamlit Community Cloud; this Cloudflare Worker provides the gunnchOS presentation shell.
Runtime: Live external Streamlit application. The interactive research app is hosted by Streamlit Community Cloud and embedded in the gunnchOS Cloudflare surface. The Cloudflare Worker itself does not execute the Python research runtime.
Evidence class: LIVE_EXTERNAL_APP. This is a simulation and research interface, not a live RAN or carrier controller. No SpectrumX or carrier endorsement is claimed.
Loading the live Streamlit app…
Research evidence
Overview
End-to-End Research Artifact
This repository contains our submission to the **SpectrumX Student Data & Algorithm Competition**. We treat the competition dataset as a mini-testbed and design an **AI-native RAN controller** that allocates radio resources under spectrum and energy constraints, with a focus on cities like **Gary, Indiana**. > **Portfolio guide:** [START_HERE](docs/START_HERE.md) · [Plain English](docs/PLAIN_ENGLISH_EXPLANATION.md) ·
Problem
### Competition Core (Phase 1) Given a 1-second IQ sample (complex-valued time series), determine whether the spectrum is **occupied** (signal present) or **unoccupied** (noise only). This is a binary classification problem with real-time inference requirements. ### Research Extension (Phase 2) How can we design AI-driven radio resource management that: - Respects spectral emission and coexistence constraints - Impro
Methods
Methods are the frozen protocol cited by results/experiments/rq2_cross_layer_continuity_summary.json. Policy means on this page are copied from that file for the train split and for three held-out policies. evidence_class is SYNTHETIC_SIM and latency_class is HOST_PROCESS_TIMING.
Native research runtimes stay on a local checkout. The interactive app above is hosted on Streamlit Community Cloud. The records in Research evidence are copied from this repository. The Cloudflare Worker does not execute the Python research runtime.
Artifacts
results/experiments/rq2_cross_layer_continuity_ablation.jsonresults/experiments/rq2_cross_layer_continuity_domain_shift.jsonresults/experiments/rq2_cross_layer_continuity_heldout.jsonresults/experiments/rq2_cross_layer_continuity_summary.jsonresults/experiments/rq2_cross_layer_continuity_train.jsonresults/experiments/rq2_fidelity_checkpoint_tiny.jsonresults/experiments/rq2_fidelity_recovery_sensitivity.jsonresults/experiments/rq2_information_equivalence_audit.jsonpaper/ARTIFACT_EVALUATION.mdpaper/CITATION_AUDIT.mdpaper/CLAIMS_TO_EVIDENCE.mdpaper/CONFERENCE_READINESS_TODOS.md
Source
Pinned source 9591d6a135ccdb37dfd1ea489e6837ec3dde288d
Limitations
- Not a live RAN controller
- No SpectrumX or carrier endorsement is claimed
About / Build status / Evidence
What is this?
Research repository for an AI-native RAN controller and equitable spectrum allocation experiments.
The research code is not executing here. This is a research surface.
Why does it exist?
It gives this repository a public web surface inside the gunnchOS ecosystem without pretending the original runtime runs on Cloudflare Workers.
What can I actually do with it today?
Launch the live AI-RAN app to use the Streamlit research interface, then inspect the repository records under Research evidence. The Cloudflare Worker does not execute the Python research runtime.
What is not finished?
- Not a live RAN controller
- No SpectrumX or carrier endorsement is claimed
STAGING_WORKER
Runtime mode: EXTERNAL_INTERACTIVE_APP. Streamlit Community Cloud hosts the Python app. This Worker hosts the gunnchOS shell. No remote model call is made by the Worker.