cover_letter / art_Y61elgKVPKg

Dear Fireworks AI Hiring Team, Fireworks AI is building the infrastructure layer that determines how fast and reliably the world's developers can deploy generative AI — and that infrastructure problem is one of the most consequential engineering challenges of this decade. My path into this space runs from hand-coding backpropagation through time in C++ at UC Berkeley in 2004, through a NeurIPS-published neural network for protein structure prediction, to building a full RL post-training workbench in 2026 that benchmarks GRPO, DPO, and ten other algorithms across TRL, VeRL, OpenRLHF, and NeMo RL. That arc is why the Forward Deployed Product Manager role at Fireworks resonates: it sits exactly at the intersection of deep ML infrastructure knowledge and direct customer engagement that I have spent 12 years building toward. **Technical and AI/ML Foundation** My technical credibility in the LLM and inference space is grounded in hands-on systems I have shipped, not just products I have managed. The RL Workbench I built covers the full RLHF/DPO pipeline across three phases — a Reward Lab for designing and A/B testing reward functions (RLVR, learned, hybrid) 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 with GPU passthrough in Docker containers. I implemented 12 RL algorithms with algorithm-specific metric profiles and standardized throughput, memory, and convergence benchmarking across frameworks — the exact kind of comparative inference and training analysis that Fireworks customers need when selecting and optimizing model serving strategies. On the 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, and statistical rigor through bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, and saturation detection. The stack — FastAPI orchestrator, TimescaleDB, Redis job queue, Next.js dashboard, Ollama — reflects the same production-grade thinking I bring to customer-facing deployments. For multi-agent orchestration, I designed and implemented OpenClaw, a multi-agent gateway framework with subagent delegation, profile management, and session switching — enabling coordinated AI agent workflows across multiple industry verticals. This gives me direct fluency in the agent architectures that Fireworks customers are increasingly building on top of LLM inference APIs. **Connecting the Arc** The thread connecting my research background, my platform infrastructure work at Intuit, and my current AI product building is a consistent focus on developer experience at the infrastructure layer — making powerful, complex systems accessible and reliable for the engineers who depend on them. Fireworks sits at exactly that layer, and the Forward Deployed PM role is the place where deep technical knowledge and customer obsession have to operate simultaneously. **Why This Role** What specifically draws me to the FDPM function at Fireworks is the mandate to translate customer technical requirements — fine-tuning workflows, latency benchmarks, API integration — into structured product definition that feeds back into the roadmap. At Intuit, I ran a comparable loop: I used SQL and BigQuery telemetry across ~20 mobile apps and 30+ product SKUs to surface developer pain points, then translated those findings into platform investments that scaled ICE engagements 275% YoY to 675M+ in FY23 and drove throughput from 6K to 50K TPS via rSocket migration. The customer-to-roadmap feedback cycle is a motion I have executed at scale, and I want to apply it in a context where the underlying technology — LLM inference, fine-tuning, function calling, multimodal models — is advancing as rapidly as it is at Fireworks. **Selected Relevant Experience** - **RL Workbench (2026):** Built 3-phase post-training platform benchmarking 12 RL algorithms (PPO, GRPO, DAPO, DPO, SimPO, IPO, KTO, ORPO, SPPO, REINFORCE, REINFORCE++, RLOO) across TRL, VeRL, OpenRLHF, and NeMo RL with GPU Docker passthrough — direct fluency in the fine-tuning and inference optimization landscape Fireworks customers navigate. - **aeval (2025–2026):** Production model evaluation platform with adversarial safety testing, contamination detection, and CI/CD regression gates — enabling the kind of rigorous model quality benchmarking that informs inference infrastructure decisions. - **Intuit ICE Platform (2021–2024):** Delivered ICE Self-Service DevPortal and GitOps configuration, reducing developer onboarding from 2–3 weeks to minutes; scaled to 675M+ engagements in FY23 across QuickBooks, TurboTax, Mint, Mailchimp, and Credit Karma. - **Intuit SDK Starter Kits:** Extended Java and Python SDK Starter Kits with scaffolding templates, build configurations (Gradle/Maven), testing frameworks, and CI/CD integration — empowering developers to reach production-ready microservices in minutes. - **Fintellect AI (2024–Present):** Architected RAG retrieval pipeline with ChromaDB vector store, multi-provider LLM orchestration (Claude, GPT-4, Gemini) with fallback routing, structured output validation, and token budget optimization — hands-on experience with the LLM API integration patterns Fireworks customers implement. - **StreamIO AI / OpenClaw (2024–Present):** Implemented multi-agent orchestration framework with gateway protocol and subagent delegation; built MCP server exposing screen capture tools to AI coding assistants — practical experience with the agent and tool-use architectures Fireworks' function-calling models enable. - **Splunk Search Orchestration (2019–2021):** Owned Go microservices for Search Service and Search Catalog; led query performance optimization achieving up to 10x improvements for a Fortune 500 beta customer — experience translating customer performance requirements into infrastructure benchmarks and roadmap decisions. **Closing** Fireworks' mission — fastest, most scalable LLM inference, built by the team that shipped PyTorch and Vertex AI — is the right foundation for the next generation of AI applications, and the Forward Deployed PM role is where I can contribute most directly: sitting with customers, understanding their inference and fine-tuning requirements at a technical level, and translating those requirements into product investments that compound into platform leadership. I would welcome the opportunity to discuss how my background maps to the problems your customers are bringing to Fireworks today. Sincerely, **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info