# NetOps Copilot

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Jinseob Kim](https://cerebralvalley.ai/u/papilion)
- **GitHub:** https://github.com/papilion/netops-copilot-demo
- **Demo video:** https://youtu.be/a_LKu-OcV9U
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/638

NetOps Copilot validates BGP/OSPF configuration changes in a Batfish-based digital twin before they are applied to a production ISP backbone. A proposed change is simulated against the current network snapshot; the verdict — ALLOW, REVIEW, or BLOCK — is decided by fixed rules over Batfish evidence only (lost routes, broken flows, new announcements). The Meta Model API (Muse Spark 1.3) never decides the verdict: it only explains the evidence in plain language and proposes a next step, which must be re-simulated before it can be trusted. The demo covers 11 scenarios: 3 on a 5-node simple lab (OSPF cost shift, peer prefix block, new customer prefix) and 8 on a 13-node, two-region advanced lab (link failure, BGP session down, LOCAL_PREF exit shift, prefix hijack, route-reflector failure, DDoS blackhole /32, RPKI invalid origin, AS_PATH prepend). Every scenario gives identical verdicts on Cisco IOS, Arista EOS, and Juniper Junos lab variants. It runs live in GitHub Codespaces — open the repo, create a codespace, and use the web UI on port 8000 with Batfish started automatically — with a recorded-replay mode as fallback, plus an MCP server (list_scenarios, validate_change, get_policy). 47 automated tests pass. All topologies, prefixes, and AS numbers are fictional (RFC 5737 documentation ranges). Built by a network engineer who operates a large ISP backbone (AS9848 / AS4670) and reviews risky changes daily: a wrong BGP policy propagates in seconds, while today's pre-checks are still manual and slow.

## More from Meta Global AI Developer Hackathon

- [Sparkl](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/635)
- [Naveenx](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/636)
- [Rankall.io](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/637)
- [Team Hades](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/639)
- [Thalo Works](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/640)
- [MetaMining & Me](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/641)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/638. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt