cover_letter / art_3ucptsmOORE

Dear Cursor Hiring Team, Cursor's mission to automate coding is not a distant aspiration — it is already reshaping how professional developers work, and the third era you are entering, where developers direct fleets of autonomous agents rather than write most code themselves, is exactly the inflection point worth building for. My path from hand-coding backpropagation through time in C++ at UC Berkeley in 2004 to building multi-agent orchestration frameworks and RL post-training workbenches in 2025 has been a continuous thread: making complex AI systems legible, reliable, and worth depending on for real work. ## Technical Foundation My AI and infrastructure work is hands-on and production-grade. At Streamio AI, I designed and built 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 not a toy: it handles real task handoff, context routing, and agent lifecycle management, which maps directly to the orchestration model Cursor needs for cloud agents competing for compute, handling retries, and recovering from errors. I also built a production RL post-training workbench covering the full RLHF/DPO pipeline — a three-phase system (Reward Lab, Playground, Arena) that benchmarks 12 algorithms (PPO, GRPO, DAPO, DPO, SimPO, and others) across TRL, VeRL, OpenRLHF, and NeMo RL with live SSE metric streaming and GPU Docker passthrough. Building that system required reasoning carefully about resource allocation, parallelism, and how to surface results in a form that lets a researcher verify output without reconstructing every training run from scratch — the same problem Cursor faces when agents return logs, video recordings, and live previews. My aeval platform adds another relevant layer: a local-first model evaluation system with FastAPI orchestration, TimescaleDB, Redis job queuing, adversarial safety testing, and CI/CD integration with automated regression detection. Instrumenting success — task completion rate, confidence scores, edge-case flagging — is something I have built from scratch, not just theorized about. On the infrastructure side, at Intuit I scaled the ICE platform to 675M+ engagements in FY23, drove a migration from 6K to 50K TPS via rSocket supporting approximately 1.5M concurrent connections at sub-25ms TP99, and led a GCP-to-AWS migration for Mailchimp's MSaaS. I understand distributed systems well enough to have real conversations about sandboxing, reliability, and resource contention — not as a spectator. ## Why This Role The specific problems Cursor is solving for cloud agents are the ones I find most interesting: how a developer hands off a task and moves on without worrying about it, how the agent's output is made legible enough that review takes minutes rather than an hour of reconstruction, and how cloud and local agents hand off to each other as developers shift contexts. These are product design problems with deep infrastructure dependencies, and they sit exactly at the intersection where I have spent the last several years. What excites me most is the artifact and review layer. Cursor's agents return logs, screenshots, video recordings, and live previews — and the challenge of defining what those artifacts contain, how they are surfaced, and how a developer verifies output without replaying an entire session is a problem I have already been solving. My AutoEval system for robot model training repurposed StreamIO's screen capture and HLS streaming pipeline to score model outputs against natural-language rubrics, reducing evaluation cycles from 72 hours to approximately 4 minutes using zero-integration screen capture and multimodal AI. The design principle — make complex agent output reviewable at a glance — is the same one Cursor needs at scale. ## Selected Relevant Experience - **OpenClaw multi-agent orchestration** — built gateway protocol, subagent delegation, profile management, and session switching enabling coordinated AI agent workflows; directly applicable to Cursor's agent provisioning and orchestration model. - **ICE Self-Service platform (Intuit)** — reduced developer onboarding from 2–3 weeks to minutes in pre-prod and under 24 hours for production; mitigated $1M+ in projected opex growth; owned DevPortal, GitOps config, and ICE Playground end-to-end. - **RL Workbench** — designed three-phase post-training platform with live metric streaming, GPU Docker passthrough, and head-to-head framework benchmarking; built the measurement and instrumentation layer from scratch. - **aeval evaluation platform** — FastAPI orchestrator, TimescaleDB, Redis job queue, CI/CD regression detection, and automated safety gates; demonstrates ability to instrument success metrics and build the measurement layer a team can act on. - **AutoEval for robot model training** — zero-integration visual evaluation system using screen capture and multimodal AI to generate structured PASS/FAIL reports with confidence scores; reduced evaluation cycles from 72 hours to ~4 minutes. - **Splunk Search Orchestration (Senior PM)** — owned Go microservices, PostgreSQL metadata service, and SPL/SPL2; delivered Scheduler Service end-to-end in approximately 4 months; achieved up to 10x query performance improvements for enterprise beta customers. - **MSaaS Drift Detection (Intuit)** — wrote Java JAR library to scan Git repos for configuration drift and built remediation roadmap; demonstrates comfort reasoning about infrastructure state and developer-facing reliability problems. ## Closing Cursor's mission — automating coding — matters because it changes what developers can accomplish, not just how fast they accomplish it. The third era you are building toward, where a developer hands off a task to a fleet of agents and returns to something reviewable rather than a wall of diffs, requires a PM who can hold the infrastructure model and the developer experience simultaneously. That is the work I have been doing, and I would welcome the opportunity to bring it to Cursor. Thank you for your consideration. --- **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info