Same backend, same database, same k6. In the first runs it looked like Embedded GlassFish was on top. When I fixed JDK, warmup, window, heap, and DB attribution, the story changed: Spring Boot ended up with the best local profile for this workload. Payara Micro was the cleanest Jakarta EE by check failures. GlassFish surprised by being viable.
May 28 2026 · 10′ · Experiments · Performance · backend
I built a reproducible lab to compare Prisma 5 against Spring Boot JdbcTemplate on the same PostgreSQL 16. What I found wasn't a winner: it was that query shape and N+1 explain almost everything, and blaming the ORM is too easy.
May 16 2026 · 5′ · Experiments · TypeScript · Performance
I took the DeepClaude repo (467 points on HN) and dropped it into my real production loop. The combination isn't simply "better than either alone" — there's a specific task regime where DeepSeek V4 Pro destroys and Claude fails, and vice versa. Here are my actual numbers.
May 04 2026 · 8′ · Experiments · TypeScript · claude code
I built a public lab with reproducible benchmarks to measure TypeScript 7 native preview against TypeScript 6 on real repos. The results are interesting, but the more useful story isn't the speedup: it's understanding when it matters, what breaks during migration, and how to test it without exposing private code.
Apr 29 2026 · 6′ · Experiments · TypeScript · Performance
I ran GPT-5.5 against my actual production prompts and compared it to GPT-4o on latency, cost, and output quality. The marketing leap doesn't match the leap in my metrics. Here are the numbers.
Apr 25 2026 · 8′ · Experiments · TypeScript · railway
Google announced two chips designed for the "agentic era." I ran my real agent workload against their published numbers. The gap between hardware marketing and what an indie dev can actually use today is brutal — and it's not a technical limitation, it's a business decision.
Apr 22 2026 · 9′ · Experiments · produccion · infraestructura
The 45% token reduction is real. What nobody measures is how much implicit context you lose along the way. I built my own benchmark and the numbers are more complicated than the headline suggests.
Apr 19 2026 · 7′ · Experiments · LLM · agentes-ia
I approved three PRs with hardcoded keys. The same models that helped write them could have caught them. N-Day-Bench measures exactly that gap — and the numbers bother me more than I expected.
Apr 14 2026 · 8′ · Reflections · devops · LLM
Architecture, real systems, product decisions and technical lessons. Everything reviewed with my own judgment before it ships.
213 articles published