cover_letter / art_ZfxNo2Ezigc

Dear Netflix CPOP Hiring Team, Netflix sits at a rare intersection: a studio, a technology platform, and a global distribution engine operating simultaneously at scale. The Content Platform Operations & Publishing team's mandate — making the Netflix experience feel personal and relevant through AI-driven promotion, localization, and distribution — is precisely the kind of problem I have spent 12 years building toward. My path from hand-coding backpropagation through time in C++ at UC Berkeley to shipping production RL post-training workbenches and multi-agent orchestration frameworks has been a consistent pursuit of the same goal: closing the gap between cutting-edge ML research and products that work at scale for real users. ## Technical and AI Foundation My ML credibility is grounded in both research and production engineering. In 2014, I published at NeurIPS on artificial neural networks for protein secondary structure prediction — work that began with a hand-coded neural network in C++ with custom BPTT in 2004 and that I rewrote in 2026 as a full PyTorch platform spanning 413 to 8 billion parameters across five architectures (feedforward, GRU, Transformer, ESM-2, multi-task), with MLflow experiment tracking, Optuna hyperparameter optimization, and FastAPI serving backed by 823 automated tests. More recently, I built **aeval**, a local-first model evaluation platform with five core eval types (factuality, reasoning, instruction-following, safety, code generation), adversarial safety testing with refusal detection, and statistical rigor including bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, and saturation detection — with CI/CD integration for regression detection and automated safety gates. The stack: FastAPI orchestrator, TimescaleDB, Redis job queue, Next.js dashboard, and Ollama for local inference. I also built an **RL post-training 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 (MPS) or CUDA; and an Arena for head-to-head framework benchmarking across TRL, VeRL, OpenRLHF, and NeMo RL with GPU passthrough in Docker containers — implementing 12 RL algorithms with standardized throughput, memory, and convergence 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 — enabling coordinated AI agent workflows across multiple industry verticals. This is the kind of agentic AI infrastructure the CPOP team will need as it automates and personalizes content operations at global scale. ## Connecting the Arc The throughline in my career is building platforms that abstract ML complexity into tools that creative, operational, and developer teams can actually use — and then scaling those platforms to production. That is exactly what the CPOP role requires. ## Why This Role What excites me most about this position is the specific challenge of translating ML capabilities into tools for artists, linguists, editors, and content executives — audiences who have deep domain expertise but are not ML practitioners. I have done this repeatedly: at Intuit, I delivered the ICE Self-Service platform that reduced developer onboarding from 2–3 weeks to minutes, and I built the Asterias declarative asset lifecycle management platform with a GraphQL API to give non-infrastructure teams control over complex platform resources. The CPOP team's work on promotional asset creation, localization, and content distribution maps directly to the class of problems I find most compelling — where the ML is sophisticated but the product surface must be invisible to the end user. ## Selected Relevant Experience - **RL Workbench (2026):** Built end-to-end post-training platform implementing 12 RL algorithms (PPO, GRPO, DAPO, DPO, SimPO, and others) with cross-framework benchmarking (TRL, VeRL, OpenRLHF, NeMo RL), live metric streaming, and GPU Docker passthrough — demonstrating hands-on ML lifecycle ownership from reward design through model evaluation. - **aeval (2025–2026):** Designed and shipped a model evaluation platform with adversarial safety testing, data contamination detection via SHA-256 hashing, and automated safety gates in CI/CD — directly applicable to the ML evaluation rigor the CPOP team requires when integrating new models into production content workflows. - **Intuit ICE Platform — 675M+ engagements, 275% YoY growth (FY23):** Led platform infrastructure scaling from 6K to 50K TPS via rSocket migration supporting ~1.5M concurrent connections with sub-25ms TP99 across QuickBooks, TurboTax, Mint, Mailchimp, and Credit Karma — evidence of shipping platform products that operate at Netflix-scale. - **ICE Self-Service DevPortal:** Delivered developer self-service platform (DevPortal, GitOps config, ICE Playground) reducing onboarding from 2–3 weeks to minutes in pre-prod and under 24 hours for production, while mitigating $1M+ in projected opex growth — 0-to-1 platform product with measurable business impact. - **Fintellect AI — RAG pipeline and multi-provider LLM orchestration:** 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 — practical experience with the ML lifecycle components the CPOP team works with daily. - **OpenClaw multi-agent orchestration:** Designed gateway protocol, subagent delegation, and session management for coordinated AI agent workflows — directly relevant to CPOP's mandate to build intelligent systems that automate content operations. - **AutoEval — Visual Evaluation for Robot Model Training (2025):** Repurposed a multimodal AI pipeline (Claude/GPT-4V) to score model outputs against natural-language rubrics, reducing evaluation cycles from 72 hours to approximately 4 minutes — an example of applying AI evaluation rigor to reduce operational latency in a production context. - **Splunk Search Orchestration — Scheduler Service delivered in ~4 months:** Owned Go microservices, PostgreSQL metadata service, and SPL/SPL2; delivered Scheduler Service end-to-end from June to October 2019, enabling scheduled search for first-party applications — demonstrating the ability to ship complex infrastructure products on aggressive timelines. ## Closing Netflix's mission to entertain the world is not a passive ambition — it requires continuous reinvention of how content is discovered, localized, and delivered to hundreds of millions of people across every culture and language. The CPOP team is building the infrastructure that makes that reinvention possible. I want to be part of that work: bringing rigorous ML product thinking, a track record of shipping platform products at scale, and the technical depth to partner credibly with applied researchers and engineers from day one. I would welcome the opportunity to discuss how my background maps to the team's priorities. Sincerely, **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info