Shahi Rahman
Global Head, Autonomous ML Iteration & Optimization (Ads) at Meta
About
I'm Shahi Rahman, currently the Global Head of Autonomous ML Iteration & Optimization at Meta. My career has been a journey through the backbone of global technology—from co-designing the Wi-Fi mesh protocols used by billions to leading the 5G cloud-native transformation at Ericsson. Today, I focus on the 'Agentic Economy,' building the autonomous ML ecosystems and agentic infrastructure that power Meta's monetization engine. I am passionate about moving beyond 'AI-assisted' tools to truly 'AI-native' operations, particularly in how we handle hyperscale reliability and incident response. Beyond my work at Meta, I serve as a board advisor for companies like Join Digital and CloudlyIO, helping them navigate the transition to AI-native architectures. I’m here to connect with fellow architects and leaders who are looking to translate complex AI innovation into measurable, large-scale commercial success.
Networking
What I can offer
- ›Strategic advising for AI-native transitions
- ›Hyperscale infrastructure expertise
- ›Insights on translating AI innovation into commercial advantage
Looking for
- ›expanding my professional network
- ›exploring mutual opportunities in AI, NetOps, and SecOps
Best fit for
Current Interests
Background
Career
Transitioned from foundational networking at Cisco to startup leadership as a CTO, then to senior leadership at Ericsson for 5G/Cloud, and finally to Meta where he leads global infrastructure and autonomous ML ecosystems.
Education
MS in Electrical Engineering from Stanford University; BS in Electrical Engineering from The University of Queensland; Cybersecurity & Business Risk Management from Harvard University (HarvardX).
Achievements
- ›Launched SEVmate, a GenAI SRE system with 42% root cause accuracy.
- ›Directed observability for 100k+ GPU clusters for Llama-3/4 training.
- ›Secured $600M+ revenue for Ericsson via Virtual Router launch.
- ›Co-designed Wi-Fi mesh protocols used by billions of devices.
- ›Enabled 80%+ model templatization for Meta's monetization engine.
Opinions
- Incident response is an information retrieval and reasoning problem, not a dashboarding problem.
- Enterprises fail by treating AI as a chatbot bolted onto runbooks instead of rearchitecting the lifecycle.
- LLMs optimize for looking capable rather than being careful and lack pre-flight check instincts.
- Strong Human-in-the-Loop (HITL) is a necessary boundary for complex LLM tasks.