cover_letter / art_6_vHZRd53IA

Dear Dropbox Hiring Team, Dropbox sits at a genuinely interesting inflection point: the company has spent two decades earning deep trust as the place people store what matters most, and is now threading AI-powered context and intelligence through that foundation. That combination — trusted content infrastructure meeting intelligent workspace experiences — is exactly the product surface I find most compelling. My own path from building developer platform infrastructure at Intuit (675M+ annual engagements, 50K TPS) to shipping multi-agent orchestration frameworks and RAG retrieval pipelines as a founder has sharpened a specific instinct: the hardest part of AI-powered products is not the model, it is the conceptual model — helping users understand what the system knows, where it got it, and why they should trust it. **Technical and AI/ML Foundation** My technical depth spans the full stack relevant to this role. At Intuit, I owned the ICE platform — a developer-facing infrastructure layer serving ~1.5M concurrent connections across QuickBooks, TurboTax, Mint, Mailchimp, and Credit Karma — and drove a rSocket migration that scaled throughput from 6K to 50K TPS with sub-25ms TP99. That work required reasoning constantly about distributed systems tradeoffs: freshness, consistency, permission boundaries, and connector reliability at scale. These are precisely the tradeoffs the Stacks JD calls out. On the AI side, I built Fintellect AI's RAG retrieval pipeline end-to-end: ChromaDB vector store, multi-provider LLM orchestration (Claude, GPT-4, Gemini) with fallback routing, structured output validation, and token budget optimization. I have direct hands-on experience with the retrieval quality, precision/recall, and source boundary questions that define trust in AI-powered workspace products. My aeval platform — built with a FastAPI orchestrator, TimescaleDB, Redis job queue, and Ollama — implements bootstrap confidence intervals, Welch's t-test, and Cohen's d effect sizing to bring statistical rigor to model evaluation, including adversarial safety testing with refusal detection. I understand what it takes to establish quality bars and experimentation frameworks for AI systems, not just as a PM abstraction but as someone who has instrumented and measured them. My NeurIPS 2014 publication on neural networks for protein structure prediction, and the 2026 rewrite of that system spanning 413 parameters to 8B (a 19-million-fold scale increase), reflects a long-standing commitment to understanding AI systems at a foundational level — from hand-coded BPTT in C++ to modern Transformer architectures with MLflow experiment tracking and Optuna HPO. **Why This Role** The Stacks PM role is fundamentally about conceptual model clarity: how do folders, Stacks, Spaces, projects, and AI chat experiences relate to each other in a user's mental model? That is a problem I have lived. At Intuit, I led an enterprise-wide Service Language Assessment across 9 languages and built Asterias, a declarative asset lifecycle management platform with a GraphQL API — both required translating complex underlying architecture into clear, navigable mental models for thousands of developers. The challenge of making the complex seem simple, while preserving the technical integrity underneath, is the thread running through my career. **Role-Specific Connection** I am particularly drawn to the work of building intelligent project context experiences that aggregate Dropbox content, third-party references, comments, tasks, activity, and AI-generated summaries into trusted workspaces. My OpenClaw multi-agent orchestration framework — built for StreamIO with gateway protocol, subagent delegation, and profile management across real estate, insurance, health, and financial markets — is a direct analog: the core challenge is identical, surfacing the right context from heterogeneous sources while maintaining clear permission and trust boundaries. I also find the onboarding and naming system work compelling; defining promotion paths and interaction patterns that help users understand *when and why* to use Stacks is exactly the kind of 0-to-1 conceptual scaffolding I have done repeatedly, most recently in launching Fintellect AI through the App Store with guided investing journeys designed for users with no prior financial literacy. **Selected Prior Experience** - **ICE Platform at Intuit:** Achieved 275% YoY growth in 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. - **ICE Self-Service DevPortal:** Delivered 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. - **RAG Retrieval Pipeline (Fintellect AI):** Architected retrieval pipeline with ChromaDB vector store, multi-provider LLM orchestration with fallback routing, structured output validation, and token budget optimization — directly applicable to retrieval quality and freshness challenges in Stacks. - **OpenClaw Multi-Agent Orchestration (StreamIO):** Implemented multi-agent framework with gateway protocol, subagent delegation, and session management across heterogeneous content domains — experience directly relevant to building context-aware AI experiences across connectors and third-party references. - **aeval Evaluation Platform:** Built local-first model evaluation platform with adversarial safety testing, refusal detection, data contamination detection, and statistical rigor (bootstrap CIs, Welch's t-test, Cohen's d) — relevant to establishing quality bars and experimentation frameworks for Stacks AI features. - **Splunk Search Orchestration:** Owned Search Service (Go microservices), Search Catalog (PostgreSQL metadata), and SPL/SPL2; led query performance optimization achieving up to 10x improvements — direct experience with search relevance, retrieval systems, and metadata architecture. - **MSaaS Drift Detection Program (Intuit):** Initiated configuration drift detection program: wrote Java JAR library to scan Git repos, partnered with Design on DevPortal UI, and built remediation roadmap — experience building trust and reliability mechanisms into platform products. **Closing** Dropbox's mission is to design a more enlightened way of working. The Stacks initiative is one of the clearest expressions of that mission I have seen: taking the content people already trust Dropbox with and making it genuinely intelligent, contextual, and collaborative — without sacrificing the simplicity and trust that earned that position. I would bring to this role a combination of deep AI/ML technical fluency, platform infrastructure experience at scale, and a track record of shipping 0-to-1 products that make complex systems navigable. I would welcome the opportunity to discuss how my background maps to the specific challenges ahead. Sincerely, **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info