cover_letter / art_VCleovf1Wis

Dear Robinhood Cortex Hiring Team, Robinhood is doing something genuinely consequential: putting sophisticated financial tools in the hands of people who have historically been locked out of them. The $124 trillion generational wealth transfer you're positioned at the center of isn't just a market opportunity — it's a structural shift in who gets to participate in building wealth. My path from hand-coding backpropagation through time in C++ at UC Berkeley in 2004 to building production RL post-training workbenches and multi-agent AI platforms today has been oriented around exactly this kind of problem: making powerful, complex systems accessible and trustworthy at scale. **Technical Foundation** My AI/ML work spans both research depth and production delivery. At the research end, I published at NeurIPS 2014 on neural networks for protein structure prediction — work that began with a hand-coded C++ neural net with custom BPTT and was rewritten in 2026 as a full PyTorch platform spanning 413 parameters to 8B (a 19-million-fold scale increase), with MLflow experiment tracking, Optuna hyperparameter optimization, and FastAPI serving across 6 Docker containers. On the evaluation side, I built **aeval** — a local-first AI model evaluation platform with 5 core eval types (factuality, reasoning, instruction-following, safety, code generation), adversarial safety testing with refusal detection, and data contamination detection via SHA-256 hashing. Statistical rigor was built in from the start: bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, and saturation detection. The platform integrates with CI/CD pipelines for regression detection and automated safety gates — precisely the kind of systematic quality infrastructure Cortex needs to ship with confidence at scale. For post-training RL, I built a 3-phase workbench covering the full RLHF/DPO pipeline: a Reward Lab for designing and A/B testing reward functions across GSM8K, MATH, HumanEval, and UltraFeedback; a Playground for real TRL-powered GRPO/DPO training with live SSE metric streaming on Apple Silicon and CUDA; and an Arena for head-to-head framework benchmarking across TRL, VeRL, OpenRLHF, and NeMo RL with GPU passthrough in Docker containers. I implemented 12 RL algorithms — PPO, GRPO, DAPO, DPO, SimPO, IPO, KTO, ORPO, SPPO, and others — with standardized throughput, memory, and convergence benchmarking. On the agentic systems 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 architected a RAG retrieval pipeline with ChromaDB vector store, multi-provider LLM orchestration (Claude, GPT-4, Gemini) with fallback routing, structured output validation, and token budget optimization for Fintellect AI, a mobile-first financial education platform I founded. **Why Cortex, Why Now** Cortex is at exactly the inflection point I find most interesting: the transition from a capable research assistant to an AI that can reason, plan, and act on behalf of customers across their entire financial life. That requires solving hard problems simultaneously — evaluation rigor, agentic reliability, regulatory navigation, and consumer trust — and doing it at the pace the AI landscape demands. My background sits at the intersection of all four. **Role-Specific Connection** The JD's emphasis on building evaluation frameworks that let the team ship with confidence maps directly to what I built with aeval — and I understand why this is load-bearing infrastructure, not a nice-to-have. The push toward agentic financial planning also resonates with my work on OpenClaw and Fintellect Agents: domain-scoped conversational agents that deliver context-aware advisory interactions require careful design of trust boundaries, fallback behavior, and output validation — exactly the challenges Cortex will face as it moves from research to advice. And having shipped AI products in financial services, I understand that regulatory navigation isn't a constraint to route around — it's a product design input. **Selected Prior Experience** - **aeval platform**: Built automated AI evaluation system with 5 eval types, adversarial safety testing, refusal detection, bootstrap confidence intervals, Welch's t-test, Cohen's d, and CI/CD regression gates — directly applicable to Cortex's quality and safety measurement needs. - **RL Workbench**: Benchmarked 12 RL algorithms (PPO, GRPO, DPO, SimPO, KTO, ORPO, and more) across TRL, VeRL, OpenRLHF, and NeMo RL with standardized throughput/memory/convergence metrics — demonstrates the technical depth to hold informed opinions on model strategy and architecture. - **Fintellect AI — RAG pipeline and Fintellect Agents**: Architected multi-provider LLM orchestration with fallback routing, ChromaDB vector store, and domain-scoped conversational agents for financial advisory interactions — directly relevant to Cortex's architecture and product direction. - **OpenClaw multi-agent orchestration**: Built production multi-agent framework with gateway protocol, subagent delegation, and session management — production experience with agentic systems at the architecture level. - **Intuit — ICE Self-Service Platform**: Delivered developer platform that reduced onboarding from 2–3 weeks to minutes, scaled throughput from 6K to 50K TPS via rSocket migration supporting ~1.5M concurrent connections with sub-25ms TP99, and achieved 275% YoY growth to 675M+ engagements in FY23 — demonstrates ability to own complex platform roadmaps and scale infrastructure under real production load. - **Intuit — ICE Presence in async chat**: Implemented AI-powered presence feature generating $480K/month in additional invoicing — evidence of finding high-impact AI moments in consumer products and shipping them with measurable business outcomes. - **NeurIPS 2014**: Published research on neural networks for protein structure prediction — establishes foundational ML credibility that informs architectural judgment, not just product intuition. **Closing** Robinhood's mission — democratizing finance for all — is the right frame for what Cortex can become. An always-on financial partner that every customer deserves but few have had access to isn't a feature; it's a structural change in who gets expert financial guidance. I've spent the last two years building the technical foundations — evaluation infrastructure, agentic orchestration, RL post-training tooling, and financial AI products — to contribute meaningfully to that mission from day one. I'd welcome the conversation. Sincerely, **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info