Field notes from production AI
Case studies and engineering deep-dives from real enterprise deployments — not the marketing version.
How Meridian Financial cut analyst research time by 64%
Filing-grade retrieval over 12 years of SEC documents with cited answers.
Helio Health: a clinical copilot field teams actually use
PHI-safe ingestion, guideline-grounded answers, and a 92% adoption rate.
The PII redaction stack we deploy by default
A reference architecture for detection, redaction, and reversible tokenization at ingestion.
Evaluating retrieval the way auditors expect
Golden datasets, regression gates, and the metrics that survive contact with executives.
Chunking strategies for technical documentation
When semantic beats fixed-size, and how to measure the difference on your corpus.
Hybrid search without the religious war
A pragmatic framework for combining lexical and vector retrieval in production.
Northcrest Legal: matter-scoped retrieval at scale
Per-matter ACLs, citation-grade answers, and reviewer workflows that hold up.
A checklist for shipping an LLM feature on a Tuesday
Evals, monitoring, fallbacks, and the rollback plan you forgot to write.
Want the unreleased pieces?
We share early drafts and internal benchmarks with our newsletter subscribers.