cover_letter / art_uRPmD9LoLX0

Dear Stripe ML Foundations and GenAI Hiring Team, Stripe's mission — increasing the GDP of the internet — is not abstract to me. It is the infrastructure layer that makes every fintech product I have built possible. When I shipped Fintellect AI's paper-trading platform on live Alpaca market data, or wired Streamio's real estate CMA pipeline to Redfin and Zillow APIs, I was building on top of the financial rails Stripe helps power. That proximity to financial infrastructure, combined with twelve years of shipping ML-powered developer platforms, is what draws me to this role. **Technical and ML Foundation** My ML work spans from first principles to production systems. In 2004 I hand-coded a backpropagation-through-time network in C++ for protein secondary structure prediction — work that was accepted at NeurIPS 2014. In 2026 I rewrote that system as a full PyTorch platform with five neural architectures (feedforward, GRU, Transformer, ESM-2, multi-task), MLflow experiment tracking, Optuna hyperparameter optimization, and FastAPI serving across a 19-million-fold parameter scale increase (413 to 8B parameters). That arc — from hand-coded BPTT to modern transformer stacks — gives me genuine fluency at every layer of the ML stack, not just the API surface. More directly relevant to Stripe's current focus: I built a production RL post-training workbench covering the full RLHF/DPO pipeline. The platform implements 12 RL algorithms (PPO, GRPO, DAPO, REINFORCE, REINFORCE++, RLOO, DPO, SimPO, IPO, KTO, ORPO, SPPO) with live SSE metric streaming on Apple Silicon (MPS) and CUDA, and a benchmarking arena for head-to-head framework comparison across TRL, VeRL, OpenRLHF, and NeMo RL with GPU passthrough in Docker containers. I also built aeval, a local-first model evaluation platform with adversarial safety testing, refusal detection, bootstrap confidence intervals, Welch's t-test, and Cohen's d effect sizing — the kind of statistical rigor that separates evaluation infrastructure from ad-hoc benchmarking. On the agentic side, I designed and shipped OpenClaw, a multi-agent orchestration framework with a gateway protocol, subagent delegation, profile management, and session switching — coordinating AI agent workflows across real estate, insurance, health, and financial-markets verticals. I also built Fintellect's RAG retrieval pipeline (ChromaDB vector store) with multi-provider LLM orchestration across Claude, GPT-4, and Gemini, with fallback routing, structured-output validation, and token-budget optimization. **From Research to Platform Scale** The bridge between my research work and this role is the Intuit chapter. As Staff PM for Developer Frameworks and Platform Infrastructure, I owned the ICE platform from 6K to 50K TPS via an rSocket migration supporting approximately 1.5M concurrent connections at sub-25ms TP99 — achieving 275% YoY engagement growth to 675M+ engagements in FY23 across QuickBooks, TurboTax, Mint, Mailchimp, and Credit Karma. I also led a company-wide Service Language Assessment across nine languages presented to the CTO, and delivered the ICE Self-Service DevPortal that compressed developer onboarding from two to three weeks down to minutes. That experience — owning both the ML application layer and the infrastructure that empowers every other team — maps directly to what Stripe's ML Foundations team is chartered to do. **Why This Role** Stripe's ML Foundations team sits at the intersection of two problems I find most interesting: incubating net-new ML applications that drive business outcomes, and building the infrastructure and tooling that lets every other product team move faster. The JD's explicit call-out of agentic capabilities and API support for agent quality and continuous improvement aligns precisely with the evaluation and orchestration infrastructure I have been building. I am particularly drawn to the cross-BU strategy dimension — working with product leaders across the company to identify where ML investment creates the most leverage is exactly the kind of work I did at Intuit when I led the enterprise-wide language assessment and the MSaaS Drift Detection program. **Selected Relevant Experience** - **RL Workbench (2026):** Built 3-phase post-training platform (Reward Lab, Playground, Arena) implementing 12 RL algorithms with live metric streaming and cross-framework benchmarking (TRL, VeRL, OpenRLHF, NeMo RL) — direct experience with the RLHF/DPO techniques underpinning modern LLM fine-tuning. - **aeval (2025–2026):** Production model evaluation platform with adversarial safety testing, statistical rigor (bootstrap CI, Welch's t-test, Cohen's d), CI/CD regression detection, and automated safety gates — the evaluation infrastructure layer Stripe's ML teams would need to maintain agent quality at scale. - **OpenClaw / Streamio AI (2024–Present):** Multi-agent orchestration framework with gateway protocol and subagent delegation, deployed across four industry verticals; demonstrates ability to design agentic architectures beyond single-model inference. - **Intuit ICE Platform (2021–2024):** Scaled to 675M+ engagements and 50K TPS; delivered self-service DevPortal reducing onboarding from weeks to minutes; achieved $480K/month in additional invoicing via ICE Presence in async chat — proven track record of platform-level impact at Stripe-relevant scale. - **Intuit SDK Starter Kits:** Extended Java and Python SDKs with scaffolding templates, build configurations (Gradle/Maven), and CI/CD integration — experience shipping developer-facing tooling that directly improves team velocity. - **Fintellect AI RAG Pipeline (2024–Present):** Multi-provider LLM orchestration with fallback routing, structured-output validation, and 13 domain-specific AI advisors on live financial data — applied GenAI in a regulated, latency-sensitive financial context. - **NeurIPS 2014 / BRAIN Platform:** Published research on neural networks for protein structure prediction; 2026 rewrite to 8B-parameter PyTorch platform with MLflow and Optuna — establishes ML credibility from first principles through modern production systems. **Closing** Stripe's mission to put the global economy within everyone's reach is one I take seriously — I have spent the last year building financial education and investing tools specifically to close the access gap for retail investors and first-generation wealth builders. Joining Stripe's ML Foundations team would let me operate at a scale where the infrastructure decisions I make ripple across millions of businesses. I would welcome the opportunity to discuss how my platform PM background, applied ML depth, and founder experience translate into impact for Stripe. Sincerely, **O. Felix Amoruwa** famoruwa@berkeley.edu · 909-731-9011 · felixamoruwa.info