Beyond 5G (B5G) networks must support diverse services such as enhanced mobile broadband (eMBB), ultra-reliable low-latency communications (URLLC), and massive machine-type communications (mMTC). Static scheduling fails under such heterogeneity, leading to poor spectral efficiency and degraded QoS. We present TRIP-B5G, a unified framework combining a lightweight Support Vector Machine (SVM) traffic classifier and a Proximal Policy Optimisation (PPO) agent for slice-aware physical resource block (PRB) allocation in Open Radio Access Networks (O-RAN). Using Key Performance Indices (KPI)-driven features (e.g., Channel Quality Indicator, buffer occupancy, throughput), the SVM classifies traffic slices, while PPO dynamically allocates PRBs with slice-specific reward functions. Implemented on srsRAN, Open5GS, and O-RAN Near-Rt RIC, TRIP-B5G achieves up to 25% higher eMBB throughput, 28% lower URLLC latency, and 15% higher mMTC access success, reclaiming more than 90% of idle PRBs. Results confirm that KPI-driven ML and RL co-design enables practical AI-native RAN control in B5G.

We present Ambient Protein Diffusion, a framework for training protein diffusion models that generates structures with unprecedented diversity and quality. State-of-the-art generative models are trained on computationally derived structures from AlphaFold2 (AF), as experimentally determined structures are relatively scarce. The resulting models are therefore limited by the quality of synthetic datasets. Since the accuracy of AF predictions degrades with increasing protein length and complexity, de novo generation of long, complex proteins remains challenging. Ambient Protein Diffusion overcomes this problem by treating low-confidence AF structures as corrupted data. Rather than simply filtering out low-quality AF structures, our method adjusts the diffusion objective for each structure based on its corruption level, allowing the model to learn from both high and low quality structures. Empirically, ambient protein diffusion yields major improvements: on proteins with 700 residues, diversity increases from 45% to 85% from the previous state-of-the-art, and designability improves from 70% to 88%.