medium · linkedin
idea post
solution architecture
Gen AI
12 min read
May 2026
Stop wiring an LLM to a Jira ticket: a solution-architect's view of test case generation
Architecture proposal for an LLM-powered test-case generator: multi-source ingestion (Jira / Confluence / Figma / TCM), context caching + summarisation, a knowledge base plus a business-skill layer, three staged human approval gates (requirement mapping → blueprint → steps), guardrails for coverage / confidence / format / dedup, episodic memory, and direct TCM import. Solution-architect thinking with Gen AI.
medium · linkedin
automation
design patterns
11 min read
May 2026
Scalable Automation: Design Patterns for Unreliable Interfaces
Six production-hardened patterns for automation that survives unstable UIs, config drift, and production unpredictability. The Scoped Action Pattern, pagination resilience, robust form filling, element waiting strategies, and the anti-patterns that silently kill test suites.
medium · linkedin
playwright
AI agents
I built a production-grade Playwright QA agent skill stack
A launch post and long-form article about the qa-agent skill: strict POM architecture, locator contracts, no static waits, production guards, audit scripts, and security-aware artifacts for AI coding agents.
pinned · medium
essay
7 min read
Pro tip: Teach your LLMs the business, not the trivia
Don't fill LLM memory with "what I ate" reminders. Use it for stable business context: business rules, product constraints, key flows, eligibility logic, decision boundaries. RAG is for fresh source-backed knowledge. Memory / skills are for durable product understanding.
linkedin · repost
4 min read
2025
We introduced AI agents to speed up QA automation
Reposted with commentary on how AI agents changed the QA automation workflow — the parts that worked, the parts that didn't, and what "AI-accelerated QA" actually looks like in a production MVNE.
essay
7 min read
Feb 2026
AI accelerates. Humans still decide.
Notes on why every agent I ship has a human-in-the-loop boundary — and why the cost calculus for autonomous AI is often worse than it looks. Reposted on LinkedIn from Pradeep Shinde's sharp satire on agents-on-agents.
article
12 min read
Jan 2026
Surviving seven upstream outages with a mock harness
What I'd do again, and what I'd never do again. The mock-at-the-edge pattern, when it works, and when it papers over a real architecture problem.
build log
6 min read
Dec 2025
A 200-line Postgres service replaced 4 hours/week
How a tiny test-data account pool became the most-loved internal tool on the QA team — and the design choices that made it sticky.
essay
10 min read
Nov 2025
Locator maintenance is the boring tax on mobile automation
The case for a multi-model self-healing framework. Why DOM heuristics alone aren't enough, and what changed when we layered a vision model in.
article
9 min read
Oct 2025
Triaging customer tickets with video + Gemini 3
The customer sent a 4-minute screen recording and a one-line Jira ticket. How a multimodal pipeline turned that into actionable repro steps with structured metadata.
// links updated as posts publish on medium · medium.com/@anmolsoin1