Software Engineer · Platform & Systems
I build the platform layers other engineers build on. Six years at Amazon: shared services at 4M TPS, privacy compliance under real regulatory pressure, and an LLM tool I shipped because on-call triage at 2 hours a session was just too painful to ignore. Outside work, I'm currently learning AV fundamentals by building evaluation tooling for autonomous driving simulators.
The decisions behind the numbers — where things broke, how I navigated them, and what I shipped.
Moving 4M TPS from HTTP to gRPC: a latency regression at 1% dialup, scope creep from a legacy Groovy module, and a deadline slip — navigated by holding the line until the root cause was fixed rather than shipping known debt.
Mid-execution on DMA standardization, a supply team demanded a source-specific bypass. I pushed for the exact revenue number first, traced the root cause to a signal-merging bug on their side, and made the $50K/week cost a bounded, time-limited decision rather than an open-ended emergency.
Privacy compliance deadline with external customer commitment collided with an auction design migration locked across four teams. Used feature flags to unblock the supply team early, extended auction timeline by one month with full alignment, shipped privacy on time.
Nobody asked me to fix the miscalibrated alarm ownership list. I used agentic tooling to inventory 250+ alarms, delegated identity team analysis to the right owners, validated 17 business-critical alarms, and eliminated 5–6 false pages per rotation cycle.
Tasked with designing the gRPC integration framework for a new Vert.x service — a technology I'd never used — from scratch. Diagnosed a subtle asyncStub → futureStub latency issue during load testing. Built in 3.5 weeks; first team onboarded in under 2 days.
Delegated majority of a Render Endemic Creative feature to a junior developer. When external timeline pressure landed on him, I ran negotiation role-plays — acting as the partner team pushing on deadline — so he could practice holding position before doing it for real. He handled the final conversations largely independently.
Building in public — systems, tools, and experiments outside of work. This section grows as I ship.
Automated evaluation framework for autonomous driving simulators — log-replay safety metrics, A* grid-world simulation, and statistical fidelity testing. Key finding: a miscalibrated generator passes every safety rule check but fails all five statistical tests. Safety metrics are necessary, not sufficient.
Let's talk
Targeting Senior SWE roles in platform engineering, AI infrastructure, or distributed systems — places where scale is real and ownership runs deep. If you're building something serious and want someone who can own it end-to-end, I'd like to hear about it.