Software Engineer · Bangalore
I read flame graphs and typography specimens with the same unhealthy attention. I think in systems — backends, color palettes, same thing to me. I build like I have to live with what I ship. Unfortunately, I do.
S. Shivam
The right abstraction compounds quietly. A good data model makes the next three years of features feel obvious. A bad one makes every feature feel like a negotiation with your past self.
Five years across healthcare, fintech, privacy, and enterprise SaaS. Different domains, same engineering instincts.
Complexity is a one-way door. Every abstraction, every indirection, every clever pattern adds a permanent tax on everyone who touches the system after you. The goal is software that a stranger could understand on their first read.
Every dependency will eventually be unavailable. Every network call will eventually time out. Every disk will eventually fill. I design around these certainties, not despite them. A system that fails legibly is better than one that fails silently.
The data model is the design. It determines what’s easy and what’s hard for years after you’ve stopped thinking about it. Finding the right boundaries — where to hide information, where to expose it — is the highest-leverage engineering work there is.
You can’t reason about what you can’t measure. A system without instrumentation isn’t a system you understand — it’s a system you’re guessing about. Monitoring isn’t an afterthought; it’s how you maintain understanding as things change.
Selected
Case studies from shipping systems that have to survive real users, real scale, and real 3 AM pages.
Built an automated space planning platform that reduced planning cycles from weeks to days.
Space planning: weeks → days
Built an ML + LLM system to automatically discover and classify sensitive data across codebases, databases, and SaaS integrations.
40% reduction in compliance audit cycles
Recent
After processing 50,000+ daily FHIR requests at Cerner, I can tell you: the spec describes an ideal world. Production describes the one where two hospitals encode blood pressure differently and DocumentReference has 47 optional fields.
The architecture behind an ML-powered system that continuously scans codebases, databases, and SaaS integrations for sensitive data.
Two million documents. One month. Zero downtime. Here's how we migrated healthcare documents from Cloudinary to AWS S3.
Growing
Short notes and seedling ideas. Some are rough. Some are evergreen. All of them are being tended to.
The right abstraction beats the right implementation. Spend disproportionate time on the data model.
Design documents are not bureaucracy — they are where the actual thinking happens.
AI should act as a force multiplier for human intent, not a replacement for human judgment.
Connect
I take on a few projects at a time. I prefer the ones where I can own the full problem — from the design doc through the production dashboard. If you’re building something that needs to work reliably at scale, I might be able to help.