cover_letter / art_7g3NKkA5npc

Dear Baseten Hiring Team, Baseten is solving one of the most consequential friction points in AI today: the gap between a model that runs on a laptop and one that serves production traffic reliably at scale. That gap costs teams weeks of engineering time and keeps capable models out of the hands of the developers who need them. I've spent the last several years on both sides of that gap — building the infrastructure that closes it at Intuit and building my own production AI systems from scratch — and the opportunity to own developer experience at Baseten is exactly where I want to apply that work. **Technical and AI/ML Foundation** My technical credibility starts in 2004, when I hand-coded a neural network in C++ with custom backpropagation through time for protein structure prediction at UC Berkeley — work that led to a NeurIPS 2014 publication. That same system I rewrote in 2026 as a full production ML platform in PyTorch, spanning 5 neural architectures (feedforward, GRU, Transformer, ESM-2, multi-task), MLflow experiment tracking, Optuna hyperparameter optimization, FastAPI serving, and Docker orchestration across 6 containers — scaling from the original 413-parameter model to 8B parameters, a 19-million-fold increase. More directly relevant to Baseten's current trajectory: I built an RL post-training workbench that benchmarks GRPO, DPO, PPO, DAPO, REINFORCE, REINFORCE++, RLOO, SimPO, IPO, KTO, ORPO, and SPPO across TRL, VeRL, OpenRLHF, and NeMo RL — with live SSE metric streaming, GPU Docker passthrough, and standardized throughput/memory/convergence benchmarking. This is the same problem space Baseten Loops is entering with its Training SDK for frontier RL workloads. I understand what developers need from that surface because I built one myself. I've also built and shipped developer-facing platform infrastructure at production scale. At Intuit, I owned the ICE platform — a developer framework serving QuickBooks, TurboTax, Mint, Mailchimp, and Credit Karma — scaling throughput from 6K to 50K TPS via rSocket migration supporting approximately 1.5M concurrent connections at sub-25ms TP99. That is the infrastructure performance envelope Baseten operates in, and I know what it takes to build the developer tooling layer on top of it. **Why This Role** Baseten's recent releases — DFlash's 3x inference speedup, sub-second image generation with Flux.2, Frontier Gateway's unified model API, and EAGLE-3 speculative decoding — signal a company that is compressing the distance between research and production at an accelerating pace. The PM role owning developer experience is the connective tissue between that infrastructure velocity and the developers who need to harness it. That is the exact role I've played before, and I want to play it here at a moment when the stakes are highest. What excites me most about this specific role is the scope: owning the full journey from first sign-up through production deployment, including CLI and SDKs, the Model Library, deployment lifecycle management, and multi-model composition via Chains. The explicit framing of the coding agent as a first-class user — not a feature — reflects a design philosophy I've already built toward. My OpenClaw multi-agent orchestration framework, with its gateway protocol and subagent delegation architecture, was built on exactly that premise. **Selected Prior Experience** - **Intuit ICE Platform:** Delivered the ICE Self-Service platform (DevPortal, GitOps config, ICE Playground), reducing developer onboarding from 2–3 weeks to minutes in pre-production and under 24 hours for production, while mitigating $1M+ in projected opex growth. Achieved 275% YoY growth in ICE engagements, scaling to 675M+ in FY23. - **Java and Python SDK Starter Kits:** Extended SDK scaffolding templates, build configurations (Gradle/Maven), testing frameworks, and CI/CD integration — enabling developers to go from zero to production-ready microservice in minutes. - **RL Post-Training Workbench:** Built a 3-phase workbench covering Reward Lab (A/B testing reward functions across GSM8K, MATH, HumanEval, UltraFeedback), a TRL-powered training Playground with live metric streaming on Apple Silicon (MPS) and CUDA, and a framework Arena for head-to-head benchmarking with GPU Docker passthrough. - **aeval — AI Model Evaluation Platform:** Built a local-first evaluation platform with 5 core eval types, adversarial safety testing with refusal detection, bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, and CI/CD integration with automated safety gates. Stack: FastAPI, TimescaleDB, Redis, Next.js, Ollama. - **Fintellect AI — Multi-Provider LLM Orchestration:** Architected a RAG retrieval pipeline with ChromaDB, multi-provider LLM routing (Claude, GPT-4, Gemini) with fallback routing, structured output validation, and token budget optimization — directly analogous to the model serving tradeoffs Baseten's customers navigate daily. - **Splunk Search Orchestration:** Owned Search Service (Go microservices), Search Catalog (PostgreSQL), and SPL/SPL2 — delivered the Scheduler Service end-to-end in approximately 4 months, and led a query performance initiative achieving up to 10x improvements in Splunk Cloud Services search for a Fortune 500 beta customer. - **Enterprise Language Assessment at Intuit:** Conducted a company-wide Service Language Assessment across 9 languages — Java, Python, Kotlin, Go, TypeScript, Scala, PHP, C++, Groovy — analyzing usage data and developer feedback to inform strategic investment decisions presented to the CTO. This is the kind of rigorous, data-grounded developer research Baseten's Voice of the Developer mandate requires. **Closing** Baseten's mission — making it effortless for any developer to go from a model on their laptop to production inference in minutes — is not a nice-to-have. It is the unlock that determines which AI companies can move fast enough to matter. I've spent 12 years building the platforms, SDKs, and tooling that make that possible, and the last two years building production AI systems myself so I understand the friction from the inside. I'd welcome the chance to bring that experience to Baseten and help define what great developer experience looks like at the frontier of inference. --- O. Felix Amoruwa famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info