Writing about retrieval, agents, and the work of finding.
Engineering deep-dives, customer stories, and the occasional opinion on where AI-native software is going. Weekly, mostly.
Permission-aware retrieval at scale, or: how we stopped leaking
Why doing ACLs at query time is harder than at index time — and why every shortcut you've heard of is wrong.
Read · 12 minCross-app entity resolution, explained
How Findola decides that "Acme" in Salesforce, "@acme" in Slack and "acme.com" in Gmail are the same customer.
Read · 8 minThe eval harness we ship to customers
Why we built our own and what we learned about benchmarking RAG systems on real corpora.
Read · 14 minHow Parallax recovered nine hours a week per consultant
A 480-person consultancy retired three knowledge-base projects and moved utilization up four points in a quarter.
Read · 6 minGlean is good. Glean is not the answer for the next 100,000 companies.
Why mid-market belongs to a different shape of product than F2000 enterprise.
Read · 9 minVerifier models: catching the failure modes that erode trust
Why a second, independent model checking the first is the cheapest hallucination reduction we've shipped.
Read · 11 minAtlas Health: HIPAA, audit log, and answers in 1.6 seconds
What deploying Findola in a regulated environment actually looks like — week by week.
Read · 7 minIntroducing Workflow Studio
From answer to action: the visual canvas for composing agentic workflows on top of Findola.
Read · 5 minHow we keep p99 latency under two seconds
Prompt caching, semantic dedup, distilled hot-path models, and the boring CDN engineering that ties it all together.
Read · 13 minSubscribe to the weekly
Engineering essays + customer notes. No promo emails. Unsubscribe in one click.