Blogs
Written from the work, not for the algorithm
Practical notes on computer vision, generative AI, and data quality - mostly about the parts that are harder than they look.
What changed in AI in 2026: 5 things that matter for real projects
Five changes from 2026 that matter if you are planning an AI project: agents in real workflows, smaller models, more AI at the edge, vision-language models, and buyers asking for proof.
Read article
Running real-time vision models at the edge: quantization, latency budgets, and zero-uplink architectures
A technical guide to deploying computer vision models on constrained on-site hardware: INT8/FP16 quantization, latency budgets, multi-object tracking through occlusion, and resilient zero-uplink architectures.
05, Oct 2026
Why a dedicated AI team beats re-hiring a vendor for every project
Lessons from our longest engagements with Sorted Tech and MiniAi Live: one team owning the full stack from labelling to edge deployment, and moving from computer vision into LLMs and agents as the technology changed.
11, Sep 2026
Why AI models fail after going live, and how to stop it
Data drift, missing evaluation checks, growing costs and no monitoring: the common reasons models break in production, and the MLOps practices that keep them working.
26, Aug 2026
RAG vs fine-tuning: which one does your business actually need?
A plain guide to choosing between retrieval-augmented generation and fine-tuning for your LLM product: what each is good at, what each costs, and how to decide before you spend money.
17, Jul 2026
Turning sports footage into insight players and venue owners can actually use
Building Athlyt’s AI sports analysis platform: one analysis serving two very different audiences. Players see what to work on, and venue owners see how their site performs.
9, Jun 2026