Qwen3.6-27B Deep Dive: Why This Mid-Size Dense Model Works So Well
A technical deep dive into Qwen3.6-27B: its hybrid architecture, long-context design, thinking preservation, and why 27B parameters can still perform at a flagship level.
The blog
Blog
Long-form, evidence-backed writing on agentic AI — newest first.
The LLM Efficiency Revolution: How 8B Models Now Outperform 70B Giants
Explore the paradigm shift in LLM development: from brute-force scaling to scientific efficiency. Discover the six key methodologies enabling smaller models to outperform their massive predecessors.
Decoding Gene Promoters: AI Cracks the Regulatory Grammar of Human DNA
A new AI model called PARM can predict and design gene promoter activity from DNA sequence alone—opening the door to programmable gene expression for therapy and regenerative medicine.
AWS Middle East Data Center Attacks: Strategic Analysis and Lessons Learned
Analysis of the March-April 2026 kinetic attacks on AWS Middle East data centers, including infrastructure damage, service disruptions, and strategic recommendations for cloud resilience in an era of hybrid warfare.
AI in Radiobiology & Radiopharmaceuticals: April 2026 Update
Comprehensive review of AI advances in theranostics dosimetry, precision radiotherapy frameworks, and radiopharmaceutical discovery including deep learning architectures, GNNs, and digital twin frameworks.
The Road Ahead: Agentic Omics in 2027 and Beyond
The final post in our Agentic Omics series synthesises two years of progress and charts the path forward: near-term breakthroughs, medium-term transformations, and the enduring challenges that will define biological AI …
Running vLLM with Qwen3.5-35B GPTQ on 4× Nvidia T4 GPUs
Comprehensive guide to running Qwen3.5-35B GPTQ Int4 on 4× Nvidia T4 16GB GPUs using vLLM with tensor parallelism. Includes architecture, configuration, performance analysis, and troubleshooting.
The Self-Driving Laboratory: Where Agents Meet Robots
The ultimate vision of agentic omics is the self-driving laboratory: AI agents that design experiments, control robotic platforms, analyse results, and iterate — with humans providing goals and oversight.
Biosecurity and Dual-Use Risks of Biological AI
AI models that can design proteins, predict pathogen evolution, or generate novel biological sequences raise real biosecurity concerns. This post takes an honest look at the dual-use risks and governance frameworks.
Open Source vs. Closed: The Battle for Biological AI
The tension between open science and commercial interests is acute in biological AI. AlphaFold 2 was open; AlphaFold 3's code was initially restricted. ESM is open; many pharma AI tools are not. This matters for …
