Articles
Page 1 of 6Practical guides on automation, ecommerce optimisation, and data engineering by Fawad Hussain Syed.
How to Benchmark a Decision Model: Lessons From Testing Jev
Six evaluation choices moved one benchmark in the same direction in a day and a half. The checks that catch them: three baselines, fitted thresholds, counted missing outputs, and validating any data you rewrite.
Jev Tested on 415 Documents: What Held Up
A day and a half testing TypeSafe's Jev decision model on 415 public disaster reports. A one-line keyword search beat it first; after four evaluation fixes its lead over Claude Haiku was 1.5 points, not 30.
How Jev Works, and How to Test a Decision Model
Jev answers typed questions with probabilities instead of text. What that means, and how to test a model like it: baselines first, fitted thresholds, AUC, calibration, guard questions, and checking data you have rewritten.
Smaller Models, Better Answers
A half-billion-parameter model can beat a frontier model on a narrow task, and the reason is not what the headline suggests. How to pick a model for your own business data, and why the evaluation matters more than the choice.
The Harness Matters More Than the Agent
Two teams running the same model get very different results, and the difference is the scaffolding around it. Sandboxes, scoped permissions, curated instructions and the feedback loop that tells an agent it is wrong.
Cognitive Debt and the Return of Boring Engineering
Thoughtworks named the thing teams have been feeling: code is arriving faster than anyone can understand it. The old practices coming back are the ones that create understanding rather than measure it.
Verification Is the Bottleneck Now
Generation got cheap and review did not. What the 2026 numbers on AI code quality actually measure, which of them you can check, and how to spend a fixed review budget on twice the code.
What to Actually Hand a Coding Agent
Coding agents are good at a narrower set of jobs than the marketing suggests and better at them than the scepticism suggests. Which tasks to delegate, how to write the brief, and why the handover matters more than the model.
Do You Still Need a Dashboard?
The case against dashboards is louder than it has ever been and thinner than it looks. What the evidence actually shows about chat, agents and semantic layers, and what a dashboard is still the only thing that does.
Can AI Build Your Dashboard Yet? What Works and What Breaks
AI can generate a dashboard from a prompt in seconds. Whether it generates a correct one is a different question, and the benchmarks answer it. What the evidence says about generated layouts, narratives, forecasts and chart choice.
Vision AI for Ecommerce: Tagging, Search, and QC That Actually Ship
Six vision AI workflows for online stores, from catalogue tagging and alt text to visual search, image compliance, and photo-verified returns, with the schemas, costs per 10,000 SKUs, and the checks that keep them honest.
Run Open LLMs Locally on Your Mac: What Works, Which Models, and What For
A practical guide to running open-weight models on Apple Silicon in 2026: the runtimes worth installing, which models fit 16, 32, and 64 GB, what you can genuinely do with them, and the pitfalls that waste a weekend.