GENESIS: AI Agents That Build & Test 6G Radio Networks

Summary

Cellular research and development is throttled by six structural processes that each consume months of manual engineering work per iteration: synthesizing new features from standards, conformance testing, hardening against field anomalies, data-driven optimization, prototyping novel capabilities, and securing the stack. While LLMs have compressed comparable work in general software engineering, they fail on Radio Access Network (RAN) use cases—hallucinating APIs, misreading specifications, and relying on simulations that break on real hardware.

GENESIS, introduced by researchers from Northeastern University and the Institute for the Wireless Internet of Things, is an agentic AI framework that directly addresses these failures. It converts intents—whether a specification clause, a telemetry anomaly, or a research hypothesis—into solutions validated with over-the-air experiments, then feeds results back into a persistent knowledge base. The framework is built on three composable primitives (agents, skills, hooks) and a knowledge layer (SYNAPSE) that doubles as ground truth and artifact repository, making capabilities compound across runs.

Key Contributions

• Identification of six structural bottlenecks in cellular R&D that each require months of manual engineering per iteration, and a framework to automate all six.
• Composable agent architecture (agents, skills, hooks) that enables flexible orchestration of autonomous R&D workflows without retraining.
• SYNAPSE knowledge layer that serves as both ground truth for agent reasoning and persistent repository for all generated artifacts, enabling compound capability growth.
• Over-the-air validation as a core primitive, overcoming the simulation-reality gap that plagues AI-generated telecom code.
• Demonstration of intent-to-validation pipeline that converts natural language inputs (spec clauses, anomalies, hypotheses) into production-grade, tested RAN code.

Implications

For Researchers

GENESIS transforms the research workflow from manual, iterative experimentation to autonomous, reproducible pipelines. Researchers can now specify a hypothesis or a new waveform design as an intent, and the framework will synthesize code, run conformance tests, and validate with over-the-air experiments—all while logging every artifact in SYNAPSE for future reuse. This dramatically accelerates the research-to-standardization cycle and enables exploration of far more design variants than previously feasible.

For Developers

The composable primitive design (agents, skills, hooks) offers a reusable pattern for building domain-specific agent systems. Developers can extend GENESIS with new skills (e.g., for specific testing protocols or optimization algorithms) without modifying the core framework. The SYNAPSE knowledge layer provides a blueprint for creating persistent, ground-truth-anchored knowledge bases that prevent hallucination and enable agents to learn from past runs.

For Users

Network operators and equipment vendors will benefit from faster deployment of new 6G features with higher confidence in interoperability and field robustness. GENESIS’s automated hardening against field anomalies and data-driven optimization means networks can be continuously improved without manual intervention. The framework’s ability to compound capabilities across runs means each deployment makes the system smarter and more capable.

References

• https://arxiv.org/abs/2605.27360v1