cover_letter / art_ySebqN_IvVo

Dear Figma AI Platform Hiring Team, Figma sits at a rare intersection: a product used by millions of designers that is now becoming the substrate through which AI generates, edits, and ships production code. That ambition — turning a canvas into a full-stack development environment powered by frontier models — is exactly the kind of platform challenge I have spent the last several years building toward. When I read about Figma Make and the Figma agent editing files directly on the canvas, I recognized the same architectural problems I have been solving hands-on: agent orchestration, eval infrastructure, context retrieval, and developer systems that other teams build on top of. **Technical and AI Foundation** My technical credibility on this role is direct, not adjacent. In 2025–2026 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 — all wired into a CI/CD pipeline with regression detection and automated safety gates. The stack is FastAPI orchestrator, TimescaleDB, Redis job queue, Next.js dashboard, and Ollama. This maps directly to Figma's "evals and continuous quality" pillar: the methodology and infrastructure that lets every AI feature improve week over week. On agent infrastructure, I built **OpenClaw**, a multi-agent orchestration framework with a gateway protocol, subagent delegation, profile management, and session switching — enabling coordinated AI workflows across multiple industry verticals inside StreamIO. I also implemented MCP SDK integration (Claude), exposing screen capture tools to AI coding assistants through a standards-based server. Figma's agent infrastructure integrates frontier models and MCP servers; I have shipped both sides of that interface. For post-training and model behavior, I built an **RL 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 running real TRL-powered GRPO/DPO training with live SSE metric streaming on Apple Silicon (MPS) and CUDA; and an Arena for head-to-head framework benchmarking (TRL, VeRL, OpenRLHF, NeMo RL) with GPU passthrough in Docker containers. I implemented 12 RL algorithms with algorithm-specific metric profiles and standardized throughput/memory/convergence benchmarking. This is the kind of systems thinking — understanding how complex components interact and improving the whole — that the AI services and evals pillars at Figma require. My research foundation goes back further: a **NeurIPS 2014** accepted paper on artificial neural networks for protein secondary structure prediction, and a 2026 rewrite of that original C++ BPTT system into a production PyTorch platform spanning 413 to 8B parameters across five architectures, with MLflow, Optuna HPO, FastAPI serving, and 823 automated tests. **Why This Role, Why Now** At Intuit I owned developer platform infrastructure at a scale most platform PMs never touch — 675M+ ICE engagements in FY23, throughput scaled from 6K to 50K TPS via rSocket migration supporting ~1.5M concurrent connections. I extended Java and Python SDK Starter Kits, delivered a self-service DevPortal with GitOps config, reduced developer onboarding from weeks to minutes, and led a Mailchimp GCP-to-AWS migration end-to-end. That experience of building the foundations other product teams ship on top of is precisely the mandate of the Figma AI Platform role. Now, as a founder building AI-native products from scratch, I have added the hands-on implementation depth — reading PRs, participating in architecture decisions, and shipping production systems — that makes a technical PM genuinely useful in architecture conversations rather than a translator on the sideline. **Role-Specific Connection** The four platform areas Figma describes — developer systems, AI services, evals and continuous quality, and search and context — are not abstractions to me. I have built working systems in each. What excites me most is the context platform problem: how Figma leverages context from inside and outside the platform while retaining enterprise trust. My RAG retrieval pipeline at Fintellect AI (ChromaDB vector store, multi-provider LLM orchestration with fallback routing, structured output validation, token budget optimization) is a direct analogue, and the enterprise trust dimension maps to the data governance work I navigated at Intuit across QuickBooks, TurboTax, and Credit Karma. The sequencing question — what to build versus buy, and how to expose platform capabilities so product teams can ship on top of them — is where I have spent the last three years making real calls, not just writing frameworks. **Selected Prior Experience** - Built **aeval** evaluation platform with 5 eval types, adversarial safety testing, statistical rigor (bootstrap CI, Welch's t-test, Cohen's d), CI/CD regression detection, and automated safety gates — FastAPI, TimescaleDB, Redis, Next.js, Ollama. - Built **OpenClaw** multi-agent orchestration framework with gateway protocol, subagent delegation, and MCP server exposing screen capture tools to AI coding assistants. - Built **RL Workbench** benchmarking 12 algorithms (PPO, GRPO, DAPO, DPO, SimPO, and others) across TRL, VeRL, OpenRLHF, and NeMo RL with GPU Docker passthrough and live SSE metric streaming. - Architected **RAG retrieval pipeline** with ChromaDB, multi-provider LLM orchestration (Claude, GPT-4, Gemini) with fallback routing, structured output validation, and token budget optimization. - Delivered **ICE Self-Service platform** at Intuit (DevPortal, GitOps config, ICE Playground), reducing developer onboarding from 2–3 weeks to minutes; achieved 275% YoY growth scaling to 675M+ engagements in FY23. - Extended **Java and Python SDK Starter Kits** with scaffolding templates, build configurations (Gradle/Maven), testing frameworks, and CI/CD integration — enabling developers to go from zero to production-ready microservice in minutes. - Owned **Search Service** (Go microservices), Search Catalog (PostgreSQL metadata service), and Splunk Processing Language (SPL/SPL2) at Splunk; delivered Scheduler Service end-to-end in ~4 months and achieved up to 10x query performance improvements for beta customers. - **NeurIPS 2014** published researcher; BRAIN platform rewrite spans 413 to 8B parameters across five neural architectures with full MLflow/Optuna/FastAPI production stack. **Closing** Figma's mission — making design accessible to all — is being redefined right now by whether the AI platform underneath Make and the Figma agent is robust enough to let every product team ship with confidence. I want to be the PM who builds that foundation: the evals that catch regressions before they reach users, the agent infrastructure that integrates frontier models reliably, the context platform that earns enterprise trust, and the developer systems that let teams go from prompt to production. I have built each of these pieces. I would welcome the chance to bring them together at Figma. Thank you for your consideration. --- **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info