Technical deep dives, industry commentary and lessons from the build. Cross-posted from LinkedIn and written for people who build or buy AI systems.
A worked example across three prompt versions: a baseline that missed edge cases, a structured version that got most things right but recommended killing a fatigued ad instead of refreshing it, and a few-shot version that fixed it. The eval framework found the failure before it reached production.
LLMs are exceptional at entity extraction. They're inconsistent at evaluation. Here's how we separate the two and why it matters for audit-grade systems.
Bills of lading, letters of credit and certificates of origin contain structured information trapped in unstructured formats. A practical look at what AI extraction can and can't do today.
Southeast Asia's sustainable finance framework is ambitious. But the gap between what regulators require and what companies can actually report remains enormous. Where does AI fit?
Shopify blocks DHL eCommerce for dietary supplement stores. Here's the workaround we built using EasyPost and PluginHive - and why platform constraints shouldn't limit your shipping strategy.
A technical walkthrough of how we orchestrate event-driven document processing workflows, with Claude handling the heavy lifting and PostgreSQL storing the results.
Most organisations can handle Scope 1 and 2 reporting with spreadsheets. Scope 3 breaks everything. A look at why supply chain emissions data needs purpose-built infrastructure.