THE BOUNDEV JOURNAL
The engineering
archive.
Earlier writing on software, hiring, and building AI. Articles retain their original URLs and publication dates; some historical offers and references are outdated.
Browse the archive
Long context vs RAG: when the big window is the wrong tool
A 1M-token window tempts teams to delete retrieval. When long context beats RAG, when it fails on cost and accuracy, and the 2026 hybrid default.
Why one AI engineer can't ship your production AI feature
A production AI feature demands six distinct skills that rarely live in one person. Here is the skill surface, why a solo AI engineer stalls at the demo, and how to plan
Your LLM eval set is too small: build it from production failures
Hand-authored LLM eval sets go stale. Here is the loop for turning real production failures into permanent regression tests.
Cut LLM input-token cost by trimming context, not output
Input tokens quietly dominate most LLM bills. How to trim conversation history, right-size RAG context, and prune system prompts to cut spend without hurting answer quali
Handling LLM rate limits (429s) in production
Your LLM feature will hit 429 rate limits under load. How to shape your own traffic with client-side limiting, backoff, queues, and key pooling so requests do not drop wh
What to monitor after you ship an LLM feature to production
The dashboard for a live LLM feature: p95 latency, token cost per tenant, refusal and error rates, and the thumbs-down signals that catch quality drift before your custom
How to roll out an AI feature safely with flags and a kill switch
A staged rollout with a real kill switch is how you get an LLM feature in front of customers without betting the account on the first prompt that goes wrong.
How to upgrade your RAG embedding model without breaking search
Swapping your RAG embedding model re-maps every vector. Here is the zero-downtime pattern to re-embed the corpus, eval the swap, and cut over safely.
Redact PII before it reaches the LLM: a gateway pattern
Put a redaction step inside your own network, in front of every outbound LLM call, so customer PII never reaches a third-party model. The reversible-tokenization pattern,
How to evaluate a senior AI engineer (a hiring rubric)
A practical 2026 rubric for interviewing senior AI engineers: the five areas that predict production skill, a 1 to 5 scoring scale, and the questions that separate builde
GO FROM THEORY TO PRACTICE
See an expert review, step by step.
Explore a sample report, a review rubric, and the evidence behind an evaluation.