cover_letter / art_9zHiUF4im0o

Dear NVIDIA AI Products Hiring Team, NVIDIA sits at the center of the most consequential infrastructure shift in computing history — not merely supplying GPUs, but defining the full stack on which enterprise AI is built, deployed, and scaled. That mission resonates directly with work I have been doing since hand-coding backpropagation through time in C++ at UC Berkeley in 2004, through publishing at NeurIPS, to building RL post-training workbenches today that benchmark GRPO, DPO, and PPO across TRL, VeRL, OpenRLHF, and NeMo RL on Apple Silicon and CUDA. The Principal Product Manager — AI Products role is the precise intersection of applied research leadership and scaled product delivery where I have spent my career. ## Technical and Research Foundation My technical credibility spans the full arc from foundational ML research to production AI infrastructure. In 2014, I published at NeurIPS on artificial neural networks for protein secondary structure prediction — work that originated from a hand-coded C++ neural network with custom BPTT in 2004 and that I rewrote in 2026 as a production PyTorch platform spanning 413 parameters to 8 billion (a 19-million-fold scale increase), with five neural architectures including ESM-2 and Transformer variants, MLflow experiment tracking, Optuna hyperparameter optimization, FastAPI serving, and 823 automated tests across six Docker containers. More recently, I built an RL post-training workbench covering the complete RLHF/DPO pipeline across three phases: a Reward Lab for designing and A/B testing reward functions (RLVR, learned, hybrid) across GSM8K, MATH, HumanEval, and UltraFeedback; a Playground running real TRL-powered GRPO and DPO training with live SSE metric streaming; and an Arena for head-to-head framework benchmarking of TRL, VeRL, OpenRLHF, and NeMo RL with GPU passthrough in Docker containers. I implemented 12 RL algorithms — PPO, GRPO, DAPO, REINFORCE, REINFORCE++, RLOO, DPO, SimPO, IPO, KTO, ORPO, and SPPO — with standardized throughput, memory, and convergence benchmarking across frameworks. This is precisely the kind of applied research-to-product translation NVIDIA describes: identifying where the field is heading, placing informed bets, and shipping systems that capture that value. On the agentic side, I built OpenClaw, a multi-agent orchestration framework with gateway protocol, subagent delegation, profile management, and session switching — enabling coordinated AI agent workflows across multiple industry verticals. I also built aeval, a local-first model evaluation platform with five core eval types, adversarial safety testing with refusal detection, bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, and CI/CD integration with automated safety gates — the kind of rigorous evaluation infrastructure that serious AI product development requires. ## The Arc That Leads Here My career has moved deliberately from deep technical work through platform infrastructure to AI product leadership — and the through-line is building systems that researchers and developers actually use at scale. NVIDIA's need for someone who can operate credibly across research scientists, enterprise customers, and cloud partners is exactly the profile I have built across twelve-plus years. ## Why This Role The specific research domains called out in the JD — agentic software development, physical AI, privacy-preserving ML, and model alignment — map directly to projects I am actively building and researching. NVIDIA's position as the infrastructure layer for cloud and AI-native customers means that product decisions here propagate across the entire industry; the opportunity to shape technical direction from research through commercialization at that scale, with NVIDIA's research teams and cloud partner relationships as leverage, is genuinely differentiated from any other role in the market. ## Selected Relevant Experience - **RL Post-Training Workbench (2026):** Built full RLHF/DPO pipeline benchmarking 12 algorithms (PPO, GRPO, DAPO, DPO, SimPO, and 7 others) across TRL, VeRL, OpenRLHF, and NeMo RL with GPU Docker passthrough — applied research translated directly into a developer-facing evaluation product. - **aeval — AI Model Evaluation Platform (2025–2026):** Designed and shipped local-first evaluation platform with adversarial safety testing, refusal detection, statistical rigor (bootstrap CIs, Welch's t-test, Cohen's d), and automated safety gates — addressing model alignment and safety evaluation at the infrastructure level. - **NeurIPS 2014 Publication:** Accepted paper on artificial neural networks for protein secondary structure prediction; original system hand-coded in C++ with custom BPTT, establishing a two-decade track record of ML research. - **Intuit — Staff PM, Developer Frameworks & Platform Infrastructure (2021–2024):** Delivered ICE Self-Service platform reducing developer onboarding from 2–3 weeks to minutes; achieved 275% YoY growth in ICE engagements scaling to 675M+ in FY23 across QuickBooks, TurboTax, Mint, Mailchimp, and Credit Karma; scaled throughput from 6K to 50K TPS via rSocket migration supporting ~1.5M concurrent connections at sub-25ms TP99. - **Intuit — Java and Python SDK Starter Kits:** Extended SDK scaffolding with build configurations, testing frameworks, and CI/CD integration — empowering developers to reach production-ready microservices in minutes; conducted enterprise-wide Service Language Assessment across 9 languages presented to CTO. - **OpenClaw Multi-Agent Orchestration (StreamIO AI, 2024–Present):** Architected and shipped multi-agent gateway framework with subagent delegation, profile management, and session switching — production agentic architecture deployed across real estate, insurance, health, and financial markets verticals. - **Splunk — Senior PM, Search Orchestration (2019–2021):** Owned Go microservices search infrastructure, SPL/SPL2 language product, and delivered Scheduler Service end-to-end in four months; led query performance optimization achieving up to 10x improvements for Fortune 500 beta customers. ## Closing NVIDIA's mission — making it possible for organizations to build, deploy, and scale AI applications with unprecedented speed and simplicity — is the mission I have been working toward from the ground up: from writing BPTT by hand in C++, to publishing ML research, to scaling developer platforms to 675M+ engagements, to building RL training infrastructure and agentic frameworks from scratch today. I would welcome the opportunity to bring that full arc to bear on NVIDIA's AI product portfolio. Thank you for your consideration. --- **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info