cover_letter / art_dOn8KEEI-08

Dear Cerebras Systems Hiring Team, Cerebras is doing something genuinely rare: rethinking the hardware substrate of AI from first principles, not iterating on legacy GPU architectures. The wafer-scale approach — delivering the compute density of dozens of GPUs on a single chip — isn't a marginal improvement; it's a platform shift. I've spent the past two years building production AI systems where inference latency was the binding constraint, and I've felt that constraint acutely. That experience is what draws me to this role. **Technical Foundation** My AI work is hands-on and spans the full stack. In 2026, I built an RL post-training workbench covering the complete RLHF/DPO pipeline: a Reward Lab for designing and A/B testing reward functions across GSM8K, MATH, HumanEval, and UltraFeedback; a Playground running real TRL-powered GRPO and DPO training with live SSE metric streaming on Apple Silicon (MPS) 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 — PPO, GRPO, DAPO, REINFORCE, REINFORCE++, RLOO, DPO, SimPO, IPO, KTO, ORPO, SPPO — with standardized throughput, memory, and convergence benchmarking across frameworks. This is the kind of work that gives me a concrete, operational understanding of what inference speed and training efficiency actually mean at the model level. On the inference side, I architected the RAG retrieval pipeline for Fintellect AI — multi-provider LLM orchestration across Claude, GPT-4, and Gemini with fallback routing, structured output validation, and token budget optimization. I also built OpenClaw, a multi-agent orchestration framework with a gateway protocol, subagent delegation, profile management, and session switching, enabling coordinated AI agent workflows across real estate, insurance, health, and financial markets verticals. These aren't toy projects; they're production systems where latency, reliability, and cost directly affect user experience. My research roots go back further. My NeurIPS 2014 paper on artificial neural networks for protein secondary structure prediction — and the original hand-coded BPTT system in C++ I built at UC Berkeley in 2004 — give me a longitudinal view of where deep learning has been and where it's going. The 2026 rewrite of that system spans 413 parameters to 8 billion, a 19-million-fold scale increase, built in PyTorch with MLflow, Optuna HPO, and FastAPI serving. **Why This Role** My arc — from ML researcher to platform PM to founder building production AI systems — maps directly to what the Strategic Verticals PM role requires: someone who can sit with a Fortune 500 enterprise or an AI-native startup, understand their technical constraints and business goals simultaneously, and translate Cerebras' latency advantages into concrete, deployable solutions. I've done that translation work repeatedly, and I know how to make it stick. What excites me specifically about this role is the combination of customer depth and product influence. The responsibility to co-architect solutions with Solutions Architects and Engineering — designing PoCs that showcase sub-millisecond inference latencies and advising on model selection and fine-tuning — is exactly the kind of work I find most valuable. And the mandate to distill customer insights into structured product feedback that shapes chip and cluster design is a feedback loop I've operated in before, at a different scale, and want to operate in again at Cerebras' level of consequence. **Selected Prior Experience** - **RL Workbench (2026):** Built 3-phase post-training platform benchmarking GRPO/DPO across TRL, VeRL, OpenRLHF, and NeMo RL; implemented 12 RL algorithms with standardized throughput/memory/convergence benchmarking — directly relevant to advising customers on model fine-tuning and inference optimization. - **Fintellect AI — RAG & Multi-Provider LLM Orchestration:** Architected production RAG pipeline with ChromaDB, multi-provider LLM orchestration (Claude, GPT-4, Gemini) with fallback routing, structured output validation, and token budget optimization — hands-on experience with the inference stack Cerebras customers are building on. - **OpenClaw Multi-Agent Framework:** Engineered multi-agent orchestration with gateway protocol and subagent delegation across real estate, insurance, health, and financial verticals — direct experience with the agentic computation use cases Cerebras' inference speed unlocks. - **Intuit — ICE Platform Scaling:** Achieved 275% YoY growth in ICE engagements, scaling to 675M+ in FY23; scaled throughput from 6K to 50K TPS via rSocket migration supporting ~1.5M concurrent connections with sub-25ms TP99 — demonstrated ability to manage platform infrastructure at enterprise scale. - **Intuit — Developer SDK & Onboarding:** Delivered ICE Self-Service platform 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 — experience building developer-facing products with measurable adoption outcomes. - **Splunk — Search Orchestration PM:** Owned Go microservices, PostgreSQL metadata service, and SPL/SPL2; delivered Scheduler Service end-to-end in ~4 months; achieved up to 10x query performance improvements for a beta Fortune 500 customer — experience owning technical product roadmaps under real customer pressure. - **NeurIPS 2014:** Published research on artificial neural networks for protein structure prediction — establishes ML research credibility that supports technical conversations with Cerebras' model lab and research institution customers. **Closing** Cerebras' mission — removing the tradeoffs between model quality, speed, and cost to unlock AI creativity and potential — is the right problem to be working on. The constraint isn't ideas; it's compute that can keep up with them. I want to be the person who helps Cerebras' most strategic customers understand what becomes possible when that constraint is lifted, and who brings that understanding back to shape the product. I'd welcome the opportunity to discuss how my background maps to what you're building. Sincerely, **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info