An AI-powered Linux Traffic Control (TC) controller that dynamically optimizes bandwidth and congestion behavior using real-time kernel telemetry collected via eBPF. The system attaches eBPF programs to networking hooks (TC/XDP/kprobes) to extract metrics such as RTT, packet loss, retransmissions, queue depth, and throughput. These metrics are streamed to a user-space decision engine implementing a lightweight Reinforcement Learning–based control policy. The policy predicts congestion trends and computes optimal rate, burst, and queue discipline adjustments. Traffic shaping parameters are applied dynamically through netlink interfaces to the Linux tcsubsystem. The architecture operates as a closed-loop feedback system with millisecond-level inference overhead and zero per-packet ML execution.