cover_letter / art_8TUt090QJt0

Dear Harvey Hiring Team, Harvey is doing something genuinely difficult: building AI that operates inside the highest-stakes knowledge work environments on earth, where a misfire isn't a bad recommendation but a document that goes to a court or a counterparty. That specificity of consequence is what makes this problem interesting — and it's what drew me here. My own path into agentic AI came through building OpenClaw, a multi-agent orchestration framework with a gateway protocol, subagent delegation, and session management across real estate, insurance, and financial markets industries — work that forced me to confront the same questions Harvey is answering at legal scale: how do you make agents trustworthy enough for professionals to stake their reputation on them? **Technical and AI Foundation** My technical credibility in this space is grounded in hands-on systems work, not adjacent familiarity. I built a full RL post-training workbench covering the complete RLHF/DPO pipeline — Reward Lab for designing and A/B testing reward functions across GSM8K, MATH, HumanEval, and UltraFeedback; a Playground running real TRL-powered GRPO and 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, REINFORCE, DPO, SimPO, KTO, and others) with standardized throughput, memory, and convergence benchmarking. This is the kind of work that gives a product leader genuine intuition about what's actually hard in training pipelines — not just what the papers say. I also built aeval, a local-first model evaluation platform with five core eval types, adversarial safety testing with refusal detection, data contamination detection via SHA-256 hashing, and statistical rigor through bootstrap confidence intervals, Welch's t-test, and Cohen's d effect size — with CI/CD integration and automated safety gates. Building evaluation infrastructure from scratch taught me that trust in AI outputs is an engineering problem, not a marketing one. That lesson applies directly to what Harvey is building. My research background adds another layer: NeurIPS 2014 published work on neural networks for protein secondary structure prediction, and the original system was a hand-coded neural network in C++ with custom backpropagation through time — written in 2004. The 2026 rewrite spans 413 parameters to 8 billion, a 19-million-fold scale increase, built in PyTorch with MLflow, Optuna, and FastAPI serving. **Why This Role** I've spent 12 years moving between deep technical execution and platform-scale product leadership — from engineering at IBM to Staff PM at Intuit, where I scaled a developer platform to 675M+ engagements in a single fiscal year. The throughline has always been: how do you build infrastructure that professionals trust enough to depend on? Harvey's Agent platform is that problem applied to legal, and the governance, audit, and ethical wall requirements aren't constraints on the product — they are the product. That framing is exactly how I think about enterprise AI. **What Excites Me About This Specific Role** The 0-to-1 nature of this work is what makes it compelling. Patterns for AI agents in legal don't exist yet — Harvey is writing them. The responsibility to define how lawyers interact with agents, how agents get created and discovered, and how the underlying platform makes both possible is the kind of problem definition work I find most energizing. The explicit acknowledgment in the JD that ethical walls, audit trails, and multi-stakeholder approval flows are core product surfaces — not afterthoughts — tells me Harvey understands what enterprise trust infrastructure actually requires. I want to be in the room where those decisions get made. **Selected Prior Experience** - **OpenClaw multi-agent orchestration (StreamIO):** Designed and implemented gateway protocol, subagent delegation, profile management, and session switching — enabling coordinated AI agent workflows across multiple professional services verticals. Direct analog to Harvey's agent creation and discovery surface. - **ICE Self-Service platform (Intuit):** Delivered DevPortal, GitOps config, and ICE Playground, reducing developer onboarding from 2–3 weeks to minutes in pre-prod and under 24 hours for production — while mitigating $1M+ in projected opex growth. Demonstrates 0-to-1 platform delivery with measurable enterprise adoption outcomes. - **275% YoY growth in ICE engagements (Intuit):** Scaled platform to 675M+ engagements 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 with sub-25ms TP99. Platform scale with real governance constraints. - **RAG retrieval pipeline (Fintellect AI):** Architected ChromaDB vector store, multi-provider LLM orchestration (Claude, GPT-4, Gemini) with fallback routing, structured output validation, and token budget optimization — applied to a regulated financial advisory context with real liability exposure. - **aeval evaluation platform:** Built adversarial safety testing, refusal detection, and automated safety gates — the kind of verification and trust infrastructure Harvey needs baked into its agent platform, not bolted on afterward. - **Enterprise Service Language Assessment (Intuit):** Conducted enterprise-wide analysis across 9 languages, synthesizing usage data and developer feedback into strategic investment decisions presented to the CTO. Demonstrates comfort operating at the intersection of technical depth and executive-level product strategy. - **Splunk Scheduler Service (Splunk):** Delivered end-to-end in approximately four months, enabling scheduled search capabilities for first-party applications — shipped to enterprise customers with real compliance and audit requirements, demoed at .conf19. **Closing** Harvey's mission — transforming how legal and professional services operate, end-to-end — is one of the few places where AI can genuinely change the leverage available to people doing consequential work. I've spent the last several years building toward exactly this intersection: agentic systems, evaluation infrastructure, enterprise platform delivery, and the governance thinking that makes all of it trustworthy. I'd welcome the chance to bring that to Harvey's Agent platform. Sincerely, **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info