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 Articles retain their original URLs and publication dates; some historical offers and references are outdated.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
Human-in-the-loop approval for AI agents that take real actions
Once an AI agent sends emails, moves money, or deletes records, accuracy is not enough. Here is how to classify actions by risk, build pause-and-resume approval, and desi
Shipping a streaming AI feature: SSE, partial output, and aborts
Streaming is the highest-leverage UX change for most LLM features, and where the edge cases hide. A practical guide to transport choice, partial-output rendering, and the
Multi-tenant RAG: how to isolate customer data and stop leaks
Shipping RAG over customer data means one job above all: tenant A must never retrieve tenant B's chunks. Here is how to pick an isolation model, enforce the boundary in c
Stop sending every request to your most expensive LLM
Most production traffic is simple. LLM model routing sends each request to the cheapest model that can answer it, cutting AI spend 40-70% without losing quality.
RAG chunking strategies that actually improve retrieval
How you chunk documents bounds everything a RAG retriever can find. What the 2026 benchmarks say about chunk size, overlap, and when semantic chunking is worth it.
Reranking in RAG: the two-stage retrieval that lifts recall
Adding a cross-encoder reranking stage to RAG lifts recall@5 from about 0.59 to 0.82 in 2026 benchmarks. Here is how the two-stage pattern works and how to size it.
How to reduce LLM hallucinations in production RAG
Hallucinations in production RAG are usually a retrieval problem, not a model problem. Here are the controls that actually cut them, in the order we add them.
How long it takes to hire a senior AI engineer in 2026
Senior AI engineers take 90 to 120 days to hire in 2026, plus notice and ramp. Here is the real timeline, why it stalls, and what every week of delay costs.
What a senior AI engineer really costs in the US in 2026
The 185,000 dollar base on the job post becomes roughly 280,000 dollars a year loaded, and closer to 390,000 in year one. Here is the line-by-line 2026 build.
Observability isn't evals: why your AI agent fails silently
Observability shows an AI agent is running. Evals show it is right. The gap between them is why agents fail silently in production, and here is how to close it.
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.