cover_letter / art_eA8R0oTZbW0

Dear Adyen Hiring Team, Adyen sits at a rare intersection: the infrastructure layer that the world's largest businesses depend on for payments, and a platform complex enough that operating it well has historically required deep expertise and significant manual effort. The shift you're describing — from merchants chasing down failed transactions to simply asking a question and getting an answer — is exactly the kind of product inflection point I've spent the last several years building toward. My work building multi-agent orchestration systems, agentic product experiences, and developer-facing platforms at scale has prepared me directly for what you're trying to accomplish. **Technical and AI Foundation** My most relevant recent work is the OpenClaw multi-agent orchestration framework I built at Streamio AI — a gateway protocol with subagent delegation, profile management, and session switching that coordinates AI agent workflows across real estate, insurance, health/dental, and financial-markets verticals. This isn't a prototype; it's a production system that handles real user workflows, enforces mutation approvals before agents take consequential actions, and routes work across specialized subagents. The design question at the center of OpenClaw — what should an agent handle autonomously, what should it confirm first, and what should it never touch — is precisely the question your JD identifies as core to this role. On the evaluation and safety side, I built aeval, a local-first AI model evaluation platform with adversarial safety testing, refusal detection, and statistical rigor (bootstrap confidence intervals, Welch's t-test, Cohen's d effect size). I also built an RL post-training workbench benchmarking GRPO, DPO, PPO, and nine other algorithms across TRL, VeRL, OpenRLHF, and NeMo RL — work that required me to go deep on how models learn from feedback signals, which directly informs how I think about quality and safety bars for agentic systems. These aren't adjacent credentials; they reflect how I think about shipping AI products responsibly. For Vantage, my AI job search platform, I shipped a streaming agent chat system (SSE) over a FastAPI backend with 40+ agent tools, enforced mutation approvals, and idempotent endpoints — a production agentic system where correctness and user trust are non-negotiable. I also built native interview prep with an automated AI judge, per-role artifact generation, and async job polling. Shipping these experiences across iOS, macOS, and web — through App Review, with Apple IAP, consent-gated data flows, and hardened network UX — taught me what it actually takes to get agentic products into users' hands at quality. **The Bridge** My arc runs from hand-coding BPTT in C++ at Berkeley in 2004, through scaling developer platforms to 675M+ engagements at Intuit, to building production multi-agent systems today. What connects those experiences is a consistent focus on reducing the distance between what someone wants done and it being done — which is exactly the mission you've articulated for this role. **Why This Role** What excites me most about this position is the specific design challenge: building an AI assistant that works on a merchant's behalf across a platform of genuine complexity — regulated, high-stakes, unforgiving of error — and making that complexity invisible to the merchant. I've shipped clear, simple experiences on top of complicated systems (multi-agent orchestration, real-time HLS pipelines, App Store–sandboxed trading copilots), and I know that simplicity is earned by going deep on the domain, not by avoiding it. The opportunity to set the quality and safety bar for merchant-facing agentic capabilities at Adyen's scale, and to build the shared foundation that lets teams across Adyen contribute capabilities, is the kind of platform-building challenge I find most compelling. **Selected Prior Experience** - Implemented OpenClaw multi-agent orchestration framework (gateway protocol, subagent delegation, profile management, session switching), coordinating AI agent workflows across real estate, insurance, health/dental, and financial-markets industries — with enforced mutation approvals before consequential actions. - Shipped Vantage streaming agent chat (SSE) over a FastAPI backend with 40+ agent tools and enforced mutation approvals; built per-role artifact generation and async job polling backed by idempotent endpoints. - Built aeval AI model evaluation platform with adversarial safety testing, refusal detection, and statistical rigor (bootstrap confidence intervals, Welch's t-test, Cohen's d); CI/CD integration with regression detection and automated safety gates. - At Intuit, achieved 275% YoY growth in ICE engagements, scaling to 675M+ 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. - Delivered ICE Self-Service platform (DevPortal, GitOps config, 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 — demonstrating the ability to build shared foundations that teams across an organization adopt. - Built Fintellect's RAG retrieval pipeline (ChromaDB vector store) with multi-provider LLM orchestration (Claude, GPT-4, Gemini), fallback routing, structured-output validation, and token-budget optimization; shipped 13 specialized AI advisors delivering guided, context-aware financial advisory. - Worked closely with telemetry and usage data (SQL, BigQuery) to prioritize developer pain points across ~20 mobile apps and 30+ product SKUs at Intuit; designed RICE-based prioritization frameworks at Splunk balancing internal partner, third-party developer, and Fortune 500 customer requirements. **Closing** Adyen's mission — enabling businesses to achieve their ambitions faster — maps directly to what I find most meaningful in product work: removing friction between intent and outcome at scale. The payment operations lead who can now ask a question instead of chasing down a failed batch, the analyst whose dispute evidence is drafted before they start — those are real improvements in how people spend their working hours, at some of the largest businesses in the world. I'd welcome the opportunity to discuss how my background building agentic systems, developer platforms, and AI-native products positions me to contribute to that mission. --- O. Felix Amoruwa famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info