Projects/Active2025 – present
NIKA
Network Incident Benchmark for AI Agents
NIKA (Network Incident Benchmark for AI Agents) is an open platform for benchmarking AI agents on realistic network troubleshooting tasks—think SWE-Bench, but for network engineers.
It deploys live emulated networks (Kathará, Containerlab), injects realistic faults, exposes MCP tools and telemetry, and evaluates agent submissions under reproducible conditions.
Highlights
- Hundreds of reproducible incidents across data center, campus, ISP, SDN, and Kubernetes scenarios
- Bring any agent framework; sandboxed tool access to CLIs, telemetry, and diagnostics
- Public website, leaderboard, and nika-bench releases for comparable results
Venues
ACM SIGCOMM’25 NGNO · arXiv preprint
Collaborators
SANDS Lab, KAUST · Politecnico di Torino collaborators
News
September 1, 2026
NIKA public leaderboard and nika-bench 0.2.0
We published nika-bench 0.2.0 and opened a public leaderboard for comparable agent runs on network troubleshooting.
July 7, 2026
NIKA project website is live
The NIKA project site is public, with overview, architecture, and links into the benchmark and code.
December 18, 2025
NIKA preprint released
We released a preprint on NIKA, a Network Incident Benchmark for AI Agents on network troubleshooting.
July 1, 2025
Paper accepted at ACM SIGCOMM 2025 NGNO Workshop
Our lightning paper on a playground for AI agents in network troubleshooting was accepted at SIGCOMM’25 NGNO.