Build it. Ship it. Operate it.

I personally built and launched a real-time AI translation app. The work includes audio sessions, authentication, backend deployment, entitlements, instrumentation and latency.

Short-lived credentials and entitlement checks keep access on the backend. Telemetry makes session and cost failures visible.

I won a global prompt-engineering championship in coding. The useful habit is practical: test model behaviour, inspect the failure and iterate in code.

This is evidence of hands-on fluency—not a claim to be a foundation-model researcher.

Connected-worker products at Tata, video intelligence at Airtel and connected safety at scale taught me to include installation, network loss, ownership and recovery in the product definition.

Employer deployments and non-public implementation details are not reproduced on this site.

Choose the funded workflow. Define its architecture and partner boundaries. Prototype the weak assumptions. Turn field exceptions into reusable capabilities and measure second-site effort.