What I do
I build AI systems that run in production, then I stress-test them until they break. That way they don’t break when it matters. Right now that means LLM agents and document-intelligence pipelines that turn thousands of unstructured safety records into decisions people act on.
Where I work
I’m the Data Systems & Cloud Infrastructure Engineer for Cal Poly’s Environmental Health & Safety and Risk Management departments. I was hired full-time after a year as their AI/ML intern. Along the way I shipped a document-intelligence app that automated fire-inspection violation tracking across campus buildings, deployed NLP risk classification and custom LLM agents on Amazon Bedrock, and delivered risk analysis that flagged missing claims exposure. All of the Cal Poly tools I’ve built are set to extend to the Solano campus next, and to Maritime after that.
Both sides of AI
What makes my work a little unusual is that I sit on both sides of it. I deploy models, and I also write the safeguards: the human-in-the-loop review gates, which mean a person has to approve the output before anything acts on it, the evaluation loops, and the risk frameworks that decide whether a model should ship at all. My senior project red-teamed maritime navigation systems against GPS spoofing and sensor-injection attacks, because I wanted to know how a trusted system fails before I trusted one myself.
How I work
I run adversarial review on my own work before I ship it, which means I attack it myself looking for the inputs that break it, and I turn every finding into a test. When a deterministic tool can do a job better than a model, meaning plain code that returns the same answer for the same input every time, I take the model out of the critical path and keep it for the parts that need language, not judgment. I would rather show a passing test suite than a demo.
Background
I studied Computer Engineering at Cal Poly, San Luis Obispo, class of 2026. Before my current role, I was Meera Services’ AI Digital Transformation Intern, where I re-architected a production billing pipeline and cut manual interventions by a third. I speak English, Gujarati, and Hindi natively, and Marathi at a full professional level.
Where I’m heading
I’m heading toward forward-deployed and AI engineering roles: sitting with the people who will use the system, building against their real constraints, and staying with it after it ships. Evaluation, red-teaming, and safety work are not a separate job for me, they are how I engineer. I write the tests and the review gates that decide whether a model should ship, and I run them on my own work first. Outside the day job, I take on freelance builds in that same spirit: AI tools for ops teams, rules engines for regulated work, and cinematic marketing sites, each one treated like a production system, not a demo. If you’re building in that space, or you’re testing AI systems the real world depends on, I want to hear from you.