Telco
14 years of carrier-grade networks: LTE radio, NFV, telco cloud, 5G standalone core.
Telco · AI · Security · Cloud-native
I'm Milind Gunjan. I've spent fourteen years building carrier-grade networks, from one of the first NFV-based VoLTE launches to cloud-native 5G cores. Now I work on the path to autonomous networks: AI that is fast, governed and secure, running on cloud-native infrastructure.
Tap an area to see how it fits the stack.
14 years of carrier-grade networks: LTE radio, NFV, telco cloud, 5G standalone core.
Inference that behaves under load, and governance for every model, tool and agent call.
Zero Trust for devices that can't run agents, and guardrails that turn intent into enforcement.
Kubernetes as the control plane for network functions, policy and AI.
Secure governance for the AI that autonomous networks will depend on, on the path to AI-native 6G: every model, tool and agent call approved, scoped and revocable. Four Kubernetes CRDs proposed to TM Forum's AI-Native Canvas working group; I architected it and built the AWS reference implementation.
Agentic Kubernetes guardrails with a quarantine mode. A tabletop I co-presented at ISC2 Security Congress 2026, with a source-available reference you can run offline.
Five live simulations of one H100 serving Llama 3.1 8B: prefill vs. decode, batching, speculative decoding, KV memory. Plus a 5-part series that walks through them.
From fault injection to safe closed-loop action: how to gate an AI agent's actions on a network with evidence. IEEE CQR 2026 at the RTC Conference, Chicago.
Your dashboard says 100%. Your users say the chatbot takes four seconds to answer. Both are right.
Every LLM request has two phases, and they are bottlenecked by completely different things.
Decode reads the weights once per step, no matter how many users share it. That is the good news hidden in the bottleneck.
If batching is where throughput comes from, memory decides how big the batch can get.
Four ideas, one chain. How they fit on one GPU, what each optimization pulls on, and what to actually watch.