cover_letter / art_5nuhqSNgy5k

Dear GitLab AI Product Management Hiring Team, GitLab sits at a rare intersection: a platform trusted by more than 50 million developers that is now redefining what it means to build software with AI agents at the center of every workflow. That mission resonates directly with the work I have been doing since founding Streamio AI — building production multi-agent orchestration systems from the ground up, making design decisions about context, memory, delegation, and coordination that have no established playbook. That hands-on experience, combined with 12+ years leading technical platform products at scale, is what draws me to the Staff PM, AI Agent Orchestration role. ## Technical and AI Foundation My AI work is not advisory — it is implemented and shipped. At Streamio AI I designed and built OpenClaw, a multi-agent orchestration framework with a gateway protocol, subagent delegation, profile management, and session switching that coordinates AI agent workflows across real estate, insurance, health/dental, and financial markets verticals. The architectural decisions I made there — how agents acquire and release context, how sessions persist across handoffs, how a gateway routes between specialized subagents — map directly to the platform capabilities GitLab is building: agent context, agent memory, and background agent behavior. To evaluate the models powering those agents, I built aeval, a local-first AI 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, and Cohen's d effect size. The stack — FastAPI orchestrator, TimescaleDB, Redis job queue, Next.js dashboard, Ollama — was designed to integrate into CI/CD pipelines with automated regression detection and safety gates. Understanding how to measure agent reliability at the platform level is prerequisite to building orchestration infrastructure that engineering teams can trust. On the RL side, I built a three-phase 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 and CUDA; and an Arena for head-to-head framework benchmarking across TRL, VeRL, OpenRLHF, and NeMo RL with GPU passthrough in Docker containers. I implemented 12 RL algorithms with algorithm-specific metric profiles and standardized throughput, memory, and convergence benchmarking. This depth of exposure to the training dynamics that shape agent behavior gives me a grounded perspective when translating model capabilities into platform product requirements. My research foundation goes back further: a NeurIPS 2014 accepted paper on artificial neural networks for protein secondary structure prediction, and the original 2004 system — a hand-coded neural network in C++ with custom backpropagation through time — rewritten in 2026 to span 413 parameters to 8 billion (a 19-million-fold scale increase) using PyTorch, MLflow, Optuna, and FastAPI serving across six Docker containers. ## Connecting the Arc The throughline from NeurIPS research to production orchestration frameworks to platform infrastructure at Intuit is a consistent pattern: I build the technical depth first, then translate it into product direction that engineering teams can execute against. That is precisely the profile this role requires — someone who can hold the platform architecture in mind while writing clear requirements, aligning stakeholders, and making prioritization calls in an area where patterns are still emerging. ## Why This Role at GitLab The specific work GitLab is doing on agent context, agent memory, and background agents is the hardest part of agentic platform design — the infrastructure layer that determines whether agents are reliable enough for developers to depend on in production workflows. Getting that layer right requires product judgment about tradeoffs that are not yet standardized across the industry. I want to work on that problem at GitLab's scale, where the decisions affect how tens of millions of developers interact with AI agents daily. The opportunity to shape the orchestration layer that enables agentic experiences across the entire DevSecOps platform — not a single feature, but the foundational capability set — is exactly the scope of work I am looking for. ## Selected Relevant Experience - **OpenClaw multi-agent orchestration framework** — designed gateway protocol, subagent delegation, profile management, and session switching enabling coordinated AI agent workflows across multiple industry verticals (Streamio AI) - **ICE Self-Service platform** — delivered DevPortal, GitOps config, and ICE Playground, reducing developer onboarding from 2–3 weeks to minutes in pre-prod and under 24 hours for production; mitigated $1M+ in projected opex growth (Intuit) - **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 (Intuit) - **Java and Python SDK Starter Kits** — extended with scaffolding templates, build configurations, testing frameworks, and CI/CD integration, enabling developers to go from zero to production-ready microservice in minutes (Intuit) - **aeval AI model evaluation platform** — built with statistical rigor (bootstrap confidence intervals, Welch's t-test, Cohen's d), adversarial safety testing, and CI/CD regression detection; designed to gate agent deployments on measurable reliability thresholds - **RL post-training workbench** — 12-algorithm benchmark platform with live metric streaming, cross-framework comparison (TRL, VeRL, OpenRLHF, NeMo RL), and GPU Docker passthrough; provides direct familiarity with the training dynamics that shape agent behavior - **Splunk Search Orchestration** — owned Search Service (Go microservices), Search Catalog (PostgreSQL metadata service), and SPL/SPL2; delivered Scheduler Service end-to-end in approximately four months and achieved up to 10x query performance improvements for a Fortune 500 beta customer ## Closing GitLab's stated mission — transforming how the world develops software — depends on getting the agent orchestration layer right. Developers will only trust AI agents in their workflows if those agents have reliable context, coherent memory, and predictable background behavior. Building the platform infrastructure that makes that trust possible is work I have been preparing for across every phase of my career, from hand-coding BPTT in C++ to shipping production orchestration frameworks today. I would welcome the opportunity to bring that foundation to GitLab. Thank you for your consideration. --- **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info