I’m a senior software engineer and system architect. I’ve spent 20+ years building software, most of it designing Ruby on Rails systems that have to keep working for a long time. These days AI is part of how I work: Claude and Cursor in my daily development, and Gemini on Vertex AI in my own projects.
class Engineer; end class Abhishek < Engineer EXPERIENCE = "20+ years" FOCUS = %i[architecture ai_assisted_dev] BASED_IN = "Pune, India" RIDES = "Royal Enfield" OPEN_TO = :good_conversationsTabendThe problems I’ve spent my career on, and the ones I happily think about on weekends.
Domain modelling, service boundaries, background processing and performance for products that need to survive a decade of change.
Major framework upgrades, dependency untangling, branch strategy and test coverage that let a team move faster, not more nervously.
Third-party APIs, legacy data sources, sync jobs, and the failure handling nobody notices until it matters.
Claude and Cursor across the whole development cycle, from design to refactoring to tests, with the same review discipline I’d apply to any other code.
AI writes a lot of first drafts in my workflow. It never gets the last word. Every line still goes through the same judgement 20+ years of building software taught me.
Talk through the design, edge cases and trade-offs before any code exists.
Scaffold, refactor and wire things up at a pace that used to take days.
Read it like a pull request from a new teammate: naming, queries, failure paths.
AI helps draft the tests. I decide what actually needs proving.
Deploy small, watch closely, and write down what I learned.
Personal builds and experiments, mostly in the space where Rails meets LLMs.
A conversational assistant on GPT-4o-mini, built end to end and written up as a full walkthrough for other Rails developers.
Turning scanned business documents into clean, validated fields using Gemini on Google Vertex AI.
Vertex AI, GeminiOne API in front of many models, with routing, fallbacks and usage metering, so an app can switch providers without rewriting itself.
Ideas I’m prototyping or planning, where AI meets the Rails problems I know best. These are concepts, labelled as such.
Reads a Gemfile.lock and deprecation logs, then drafts a step-by-step upgrade plan with a risk rating for every gem.
Plain-English questions about a large Rails app, answered from the code itself, with links to the exact files and lines.
A first-pass reviewer that catches N+1 queries, missing indexes and unsafe migrations before a human spends time on the PR.
Interested in one of these? I’m always happy to talk ideas through.
I write what I wish I’d found when I started: practical, tested, and without the hype.
The road and the trail teach what good architecture does: pack light, respect the conditions, and always know your way back.
I ride with the Roadshakers community.
If you’re untangling an ageing Rails codebase, planning a big upgrade, or bringing AI tools into how your team builds, I’m always glad to talk it through. I read every message.
Happy to talk about architecture reviews, Rails upgrades, AI-assisted development, mentoring, talks and writing.