cover_letter / art_Gw-Rtr63r1w

Dear Asana Hiring Team, Asana sits at a rare intersection: a platform trusted by millions of teams to coordinate their most important work, now extending that coordination layer to AI agents that can plan, execute, and act on behalf of those teams. That mission — making agents reliable, safe, and genuinely useful inside a work graph — is precisely the problem I have been building toward. When I architected the OpenClaw multi-agent orchestration framework at Streamio AI, handling gateway protocols, subagent delegation, profile management, and session switching across four industry verticals, I encountered firsthand how quickly agent reliability and trust become the central product problem once the demo works. **Technical and AI Foundation** My AI work spans from first principles to production systems. In 2004 I hand-coded backpropagation through time in C++ for protein structure prediction at UC Berkeley — work that led to a NeurIPS 2014 accepted paper. In 2025–2026 I rebuilt that system as a full PyTorch platform spanning 413 parameters to 8B, with MLflow experiment tracking, Optuna hyperparameter optimization, and FastAPI serving across six Docker containers. On the agent and evaluation side, 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, bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, and CI/CD regression gates. I also built an RL post-training workbench that benchmarks 12 algorithms (PPO, GRPO, DAPO, DPO, SimPO, and seven others) across TRL, VeRL, OpenRLHF, and NeMo RL with live SSE metric streaming and GPU Docker passthrough — giving me direct, hands-on intuition for how model behavior changes under different training regimes and reward functions. At Intuit, I owned developer platform infrastructure at scale: 675M+ ICE engagements in FY23, throughput scaled from 6K to 50K TPS via rSocket migration supporting ~1.5M concurrent connections at sub-25ms TP99. I drove a 275% YoY engagement growth trajectory while managing SDK tooling, GitOps configuration, and a declarative asset lifecycle platform with a GraphQL API — the kind of platform-scale thinking that translates directly to building agents that operate reliably across a complex work graph. **Why This Role** The arc from NeurIPS researcher to multi-agent framework builder to platform PM at scale leads naturally to the Core Agents team at Asana. I have spent the last year building the exact systems this role requires — orchestration, evaluation, reliability monitoring — and I want to do it at the scale and mission impact that only Asana's work graph can provide. What excites me specifically about this role is the combination of three hard problems in one seat: building agents that take dependable action across a richly structured work graph, designing the evaluation and red-teaming systems that make trust measurable, and shaping the human-agent handoff experience so that teams can direct agents without losing confidence or control. The work graph is a genuinely differentiated context for agents — tasks, projects, dependencies, and team memory are already structured — and that creates both a powerful reasoning substrate and a high bar for reliability that I find technically compelling. **Selected Relevant Experience** - **OpenClaw multi-agent orchestration (Streamio AI):** Designed and implemented gateway protocol, subagent delegation, profile management, and session switching — coordinating AI agent workflows across real estate, insurance, health/dental, and financial-markets verticals. - **aeval evaluation platform:** Built adversarial safety testing with refusal detection, statistical rigor (bootstrap CIs, effect sizes), and automated CI/CD regression gates — directly analogous to the evaluations, red-teaming, and monitoring systems called out in this role. - **RL Workbench:** Benchmarked 12 RL algorithms across four frameworks with standardized throughput, memory, and convergence metrics; built cross-tab workflow lineage tracking and live SSE metric streaming — hands-on fluency with model behavior and its trade-offs. - **Vantage AI platform (Streamio AI):** Shipped a FastAPI backend with 40+ agent tools, enforced mutation approvals, streaming agent chat via SSE, and idempotent endpoints — production agentic architecture with safety guardrails built in. - **Intuit ICE platform:** Delivered self-service developer platform reducing onboarding from 2–3 weeks to minutes, mitigating $1M+ projected opex growth; scaled to 675M+ engagements with sub-25ms TP99 — platform infrastructure at enterprise scale. - **Intuit SDK Starter Kits:** Extended Java and Python SDKs with scaffolding templates, CI/CD integration, and testing frameworks, enabling developers to reach production-ready microservices in minutes — developer-facing tooling and integration experience. - **Splunk Search Orchestration:** Owned Go microservices, PostgreSQL metadata service, and SPL/SPL2; delivered Scheduler Service end-to-end in four months and achieved up to 10x query performance improvements for a Fortune 500 beta customer. **Closing** Asana's mission — helping teams achieve their most important goals — becomes meaningfully more powerful when agents can be trusted to act as reliable participants in that coordination layer, not just assistants that suggest. Building that trust through rigorous evaluation, dependable execution, and intuitive human-agent collaboration is work I have been converging on across every role in my career. I would welcome the opportunity to bring that foundation to the Core Agents team. Thank you for your consideration. O. Felix Amoruwa famoruwa@berkeley.edu · 909-731-9011 · felixamoruwa.info