cover_letter / art_EGpulFYka0U

Dear Plaid Hiring Team, Plaid sits at a rare intersection: financial infrastructure that is simultaneously developer-facing, consumer-critical, and operating at a scale where the quality of the underlying intelligence layer directly determines whether millions of people gain or lose access to healthier financial lives. That mission — unlocking financial freedom through open finance — is one I have been building toward from two directions at once: as a platform PM who scaled developer infrastructure to 675M+ engagements at Intuit, and as a founder who has spent the past year building AI-native financial products from the ground up. ## Technical and AI Foundation My AI work is not recent or surface-level. In 2004, I hand-coded a backpropagation-through-time neural network in C++ for protein structure prediction — work that was accepted at NeurIPS 2014 after further development at Lawrence Berkeley National Laboratory. In 2026, I rewrote that same system as a production PyTorch platform spanning five architectures (feedforward, GRU, Transformer, ESM-2, multi-task), 413 to 8B parameters, MLflow experiment tracking, Optuna hyperparameter optimization, FastAPI serving, and 823 automated tests across six Docker containers. That 19-million-fold scale increase in a single project captures something real about how I approach AI systems: I care about the full stack, from the math to the infrastructure to the API surface. More directly relevant to Plaid's AI Foundations mandate, 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. The stack — FastAPI orchestrator, TimescaleDB, Redis job queue, Next.js dashboard, Ollama — was designed to support CI/CD integration with regression detection and automated safety gates. Building a trustworthy evaluation layer before deploying models into high-stakes workflows is exactly the kind of responsible AI practice Plaid's regulated environment demands. 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) and 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 (PPO, GRPO, DAPO, DPO, SimPO, KTO, and others) with standardized throughput, memory, and convergence benchmarking gives me a working mental model of the tradeoffs that matter when selecting and deploying models in production — not just reading about them. At Fintellect AI, I architected a RAG retrieval pipeline with ChromaDB vector store, multi-provider LLM orchestration across Claude, GPT-4, and Gemini with fallback routing, structured output validation, and token budget optimization — applied directly to financial advisory interactions where accuracy and trust are non-negotiable. ## Why This Role My platform PM career has been defined by building the infrastructure layer that other teams build on top of. At Intuit, that meant owning developer frameworks, SDK starter kits, and the ICE platform that scaled to 675M+ engagements across QuickBooks, TurboTax, Mint, Mailchimp, and Credit Karma. The work Plaid's AI Foundations team is doing — building core embeddings, representation learning, and model evaluation pipelines that power smarter financial experiences across the entire ecosystem — is the same class of problem: get the foundation right, and every product built on top of it improves. Get it wrong, and the blast radius is enormous. What specifically draws me to this role is the combination of platform primitives and applied AI in a regulated, high-trust domain. Financial data is among the most sensitive data that exists. Building AI systems that are reliable, transparent, and auditable in that context requires both deep ML fluency and genuine respect for the stakes — both of which I bring. ## Selected Prior Experience - **RL Workbench (2026):** Built 3-phase post-training platform benchmarking GRPO/DPO across TRL, VeRL, OpenRLHF, and NeMo RL; implemented 12 RL algorithms with standardized throughput/memory/convergence metrics and cross-tab workflow lineage tracking. - **aeval (2025–2026):** Built model evaluation platform with adversarial safety testing, refusal detection, data contamination detection via SHA-256 hashing, and statistical rigor (bootstrap CI, Welch's t-test, Cohen's d); CI/CD integration with automated safety gates. - **Fintellect AI (2024–Present):** Architected RAG pipeline with ChromaDB, multi-provider LLM orchestration (Claude/GPT-4/Gemini) with fallback routing, structured output validation, and token budget optimization for a financial advisory platform. - **Intuit — ICE Platform (2021–2024):** Delivered ICE Self-Service platform (DevPortal, GitOps config, ICE Playground), reducing developer onboarding from 2–3 weeks to minutes; achieved 275% YoY growth in engagements, scaling to 675M+ in FY23; scaled throughput from 6K to 50K TPS via rSocket migration supporting ~1.5M concurrent connections with sub-25ms TP99. - **Intuit — SDK & Developer Tooling (2021–2024):** Extended Java and Python SDK Starter Kits with scaffolding templates, build configurations (Gradle/Maven), testing frameworks, and CI/CD integration; conducted enterprise-wide Service Language Assessment across 9 languages presented to CTO. - **Splunk — Search Orchestration (2019–2021):** Owned Search Service (Go microservices), Search Catalog (PostgreSQL metadata service), and SPL/SPL2; delivered Scheduler Service end-to-end in ~4 months; led query performance optimization achieving up to 10x improvements for beta Fortune 500 customer. - **NeurIPS 2014:** Published research on artificial neural networks for protein secondary structure prediction; original system hand-coded in C++ with custom BPTT; 2026 rewrite spans 413 to 8B parameters across five architectures. ## Closing Plaid's mission to unlock financial freedom for everyone is not a tagline — it is a systems design problem. The intelligence layer you are building will determine whether the financial products sitting on top of Plaid's network are adaptive and trustworthy or brittle and opaque. I have spent 12 years building the infrastructure that developers depend on, and the past two years going deep on the AI systems that will define the next generation of those platforms. I would welcome the opportunity to bring both to Plaid. Thank you for your consideration. --- **O. Felix Amoruwa** famoruwa@berkeley.edu · 909-731-9011 · felixamoruwa.info