<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>ACJ Labs</title><description>ACJ Labs builds products and infrastructure for agentic commerce — focused, inspectable tools for catalog data, tax-aware receipts, and more. Free tier, no card.</description><link>https://acjlabs.com/</link><language>en-us</language><item><title>A wrong ABN and a smudged one look the same to OCR. We made sure ours doesn&apos;t guess.</title><link>https://acjlabs.com/blog/abn-checksum-verbatim/</link><guid isPermaLink="true">https://acjlabs.com/blog/abn-checksum-verbatim/</guid><description>Most extraction tools quietly &apos;fix&apos; a business number that doesn&apos;t check out. We added a fixture specifically to catch that — and confirmed the extractor transcribes what&apos;s printed, checksum-invalid or not.</description><pubDate>Sat, 25 Jul 2026 00:00:00 GMT</pubDate><category>Tax</category><category>Engineering</category><author>ACJ Labs</author></item><item><title>Build vs. buy: should your bookkeeping tool parse receipts with a raw LLM call?</title><link>https://acjlabs.com/blog/build-vs-buy-receipt-extraction/</link><guid isPermaLink="true">https://acjlabs.com/blog/build-vs-buy-receipt-extraction/</guid><description>A vision-model call gets you 80% of receipt extraction in an afternoon. Here&apos;s an honest look at the other 20% — AU GST rules, ABN checksums, schema stability — and when it&apos;s worth paying for versus building yourself.</description><pubDate>Sat, 25 Jul 2026 00:00:00 GMT</pubDate><category>Engineering</category><category>Tax</category><author>ACJ Labs</author></item><item><title>How accurate is &quot;just ask the model to categorize it&quot;? We measured — 1–2 out of 12.</title><link>https://acjlabs.com/blog/llm-taxonomy-accuracy-measured/</link><guid isPermaLink="true">https://acjlabs.com/blog/llm-taxonomy-accuracy-measured/</guid><description>Zero-shot LLM product classification into the Google taxonomy scored 1–2/12 exact matches on our test set. Grounding the model in the real taxonomy first took it to 22/24. Here&apos;s what actually changed, measured against a live API.</description><pubDate>Sat, 25 Jul 2026 00:00:00 GMT</pubDate><category>Engineering</category><author>ACJ Labs</author></item><item><title>The GST rule your receipt scanner doesn&apos;t know exists</title><link>https://acjlabs.com/blog/gst-entertainment-rule/</link><guid isPermaLink="true">https://acjlabs.com/blog/gst-entertainment-rule/</guid><description>Entertainment expenses have their input tax credit denied by default under Australian GST rules — and nothing printed on the receipt tells you that.</description><pubDate>Thu, 23 Jul 2026 00:00:00 GMT</pubDate><category>Tax</category><category>Engineering</category><author>ACJ Labs</author></item><item><title>AI product classifiers hand you category IDs that don&apos;t exist. Here&apos;s how we stopped ours.</title><link>https://acjlabs.com/blog/parka-category-id/</link><guid isPermaLink="true">https://acjlabs.com/blog/parka-category-id/</guid><description>A capable vision model labelled a winter parka as &apos;Shirts &amp; Tops&apos; — confidently. The fix wasn&apos;t a better prompt: ground every classification in the real taxonomy data.</description><pubDate>Thu, 23 Jul 2026 00:00:00 GMT</pubDate><category>Engineering</category><author>ACJ Labs</author></item></channel></rss>