The predominant use of wireless access networks is for media streaming applications, but current networks treat all packets the same and can’t determine which clients need service most urgently; with widespread software reconfigurability in networking devices, agile control policies can now be applied in access networks, requiring a system design that enables configuration, measures application performance impact (Quality of Experience), and adaptively selects new configurations, essentially a Markov Decision Process with unknown parameters; the goal is to develop QFlow, a platform to instantiate this loop and various control policies, using video streaming over YouTube as a use case, focusing on priority queueing and determining client assignments per queue each decision period; policies are developed using both model-based and model-free reinforcement learning, an auction-based system for priority service bidding, and a structured index-based policy, with experiments showing these policies on QFlow select the right clients for prioritization in high-load scenarios, outperforming known solutions with over 25% improvement in QoE and achieving a perfect QoE score of 5 over 85% of the time.