Skip to content
Insights

Notes from building AI in production

Practical writing on retrieval, evaluation, agents and automation, from engineers who run these systems.

Generative AIAug 18, 2026·7 min read

Why most RAG demos die in production, and the five things that fix it

A retrieval demo over ten PDFs is easy. A cited, permission-aware assistant over a million documents that stays accurate after every model change is engineering. Here is what changes.

Read insight
EngineeringJul 29, 2026·6 min read

Eval-driven development: the missing discipline in LLM applications

Unit tests told us when code broke. LLM features need the equivalent: golden sets, calibrated judges and regression gates. Here is how we set it up in a week.

Read insight
Search & RetrievalJul 8, 2026·5 min read

Hybrid search with RRF: why BM25 still matters in the age of embeddings

Keyword search never went away. Fusing it with dense retrieval fixes the recall problems that embeddings alone create on niche queries, with a surprisingly small amount of code.

Read insight
AutomationJun 20, 2026·5 min read

WhatsApp automation for D2C brands: what actually moves revenue

Open rates near 98% are real, but so is the fatigue from careless broadcasting. The brands that win use WhatsApp as a service channel first and a campaign channel second.

Read insight

Let's build intelligent systems that drive growth

Tachyon is the engineering partner for teams that need AI in production, not in a deck. Start with a free 60-minute discovery call.

Book a discovery callSend us a brief