<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>ai-augmented-se on Dhaval Shah</title><link>https://www.dhaval-shah.com/tags/ai-augmented-se/</link><description>Recent content in ai-augmented-se on Dhaval Shah</description><generator>Hugo -- gohugo.io</generator><lastBuildDate>Thu, 13 Aug 2026 01:00:50 +0000</lastBuildDate><atom:link href="https://www.dhaval-shah.com/tags/ai-augmented-se/index.xml" rel="self" type="application/rss+xml"/><item><title>Three Azure Cost Leaks - And the Analysis Process That Found Them</title><link>https://www.dhaval-shah.com/finops-azure-ai-review/</link><pubDate>Thu, 13 Aug 2026 01:00:50 +0000</pubDate><guid>https://www.dhaval-shah.com/finops-azure-ai-review/</guid><description>Background This is the fifth post in a series on AI-augmented software engineering across the disciplines that matter most for production grade enterprise systems. The earlier posts covered ground that's probably more familiar with software engineering fraternity:
Three Fintech Architecture Post-Mortems The GC Summary Report Wasn't Wrong Same JSON Storage Problem, Different Database A Black Friday Incident That Took 9 Days to Resolve This post covers different ground - Cloud Cost Optimization.</description></item><item><title>A Black Friday Incident Took 9 Days to Resolve - Here's the Process That Would Have Changed That</title><link>https://www.dhaval-shah.com/sre-gc-ai-review/</link><pubDate>Sat, 25 Jul 2026 01:00:50 +0000</pubDate><guid>https://www.dhaval-shah.com/sre-gc-ai-review/</guid><description>Background An older post on this blog covered the technical details of this incident - the GC types, the flags, the before-and-after metrics. This post covers something different: the process of how the investigation ran, where it lost time, and which decisions (with more structured approach) would have changed.
The incident On Black Friday, an online checkout platform running at over 1000 TPS. CPU spiking to 100%, application crashing. Restarted every 12 hours to keep the business running while the team investigated.</description></item><item><title>Same JSON Storage Problem, Different Database - What Postgres Does Differently</title><link>https://www.dhaval-shah.com/db-optimization-ai-review/</link><pubDate>Mon, 06 Jul 2026 01:00:50 +0000</pubDate><guid>https://www.dhaval-shah.com/db-optimization-ai-review/</guid><description>Background My earlier article on AI augmented Fintech post-mortem review walked through an Oracle JSON storage problem: a payment transaction table stored as JSON, a query that needed to filter on a nested field, and the chain of fixes - that the team eventually needed.
Out of curiosity - a reasonable question came out of that post: does Postgres have the same problem?
The honest answer is partly yes, partly no.</description></item><item><title>The GC Summary Report Wasn't Wrong - It Just Wasn't Complete</title><link>https://www.dhaval-shah.com/gc-comparison-ai-review/</link><pubDate>Sun, 21 Jun 2026 02:00:50 +0000</pubDate><guid>https://www.dhaval-shah.com/gc-comparison-ai-review/</guid><description>Background My old article compared Virtual Thread based implementations of Spring Core Reactor and JDK 21, building on the comparative analysis before it. Both articles measured the usual things - total processing time, memory footprint, GC pauses, CPU time - and both relied on the summary report generated by GCEasy.
The conclusion from that data was straightforward: JDK based implementation requires more GC activity and a heavier memory footprint, but this does not have any significant impact on application performance.</description></item><item><title>Three Fintech Architecture Post-Mortems - What AI-Augmented Review Would Have Caught</title><link>https://www.dhaval-shah.com/fintech-post-mortem-ai-review/</link><pubDate>Thu, 11 Jun 2026 02:00:50 +0000</pubDate><guid>https://www.dhaval-shah.com/fintech-post-mortem-ai-review/</guid><description>Background Architecture decisions rarely break at design time. They break a year later, under double the load, when an unwritten assumption proves false. I’ve seen this play out in synchronous coupling, storage models, and service splits - while architecting and designing FinTech platforms. That’s why I now run every major design decision through an AI-augmented review process.
What follows are three post‑mortems from real payment systems - and the simple prompts that would have flagged each failure before it hit production.</description></item></channel></rss>