cover_letter / art_xO_trcE9TkM

Dear Inflection AI Hiring Team, Inflection AI is doing something genuinely rare: building AI that prioritizes emotional intelligence alongside raw capability, treating empathy as a first-class engineering concern rather than a marketing afterthought. That mission resonates with me directly — when I built Fintellect AI's domain-specific conversational agents, I found that the most meaningful product decisions weren't about model accuracy alone, but about how the AI *felt* to a retail investor navigating real financial anxiety. That experience sharpened my conviction that the next frontier in consumer AI isn't just smarter models — it's more human ones. **Technical and AI/ML Foundation** My technical grounding runs from first principles to production systems. In 2004, I hand-coded a neural network in C++ with custom backpropagation through time (BPTT) for protein structure prediction — work that was accepted at NeurIPS 2014. In 2026, I rewrote that same system in PyTorch spanning 413 parameters to 8B (a 19-million-fold scale increase), with MLflow experiment tracking, Optuna hyperparameter optimization, and FastAPI serving across six Docker containers. More directly relevant to Inflection's stack: I built a full RL post-training workbench covering the complete RLHF/DPO pipeline — 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 and CUDA; and an Arena for head-to-head 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, and SPPO — with standardized throughput, memory, and convergence benchmarking across frameworks. This is the kind of technical depth that lets me engage substantively with ML researchers on fine-tuning trade-offs, not just relay requirements. 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, bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, and CI/CD integration with automated safety gates. I understand what it means to measure model quality rigorously, not just qualitatively. **Why This Role, Why Now** My career arc — from computational engineering at Berkeley to NeurIPS research, through platform infrastructure at Intuit and Splunk, to founding two AI-native products — has been building toward exactly this intersection: consumer-scale AI products that require both deep technical fluency and genuine product craft. The Senior PM role on Consumer AI & Agents at Inflection is where those threads converge. What specifically excites me about this role is Inflection's focus on memory, profile, and journeys as first-class system components — not bolted-on features. Building Pi's constrained inference and personalization systems at scale, while maintaining the emotional coherence that makes Pi distinctive, is a genuinely hard product problem. I've worked on analogous challenges: at Intuit, I scaled the ICE platform to 675M+ engagements in FY23 across QuickBooks, TurboTax, Mint, Mailchimp, and Credit Karma, with throughput scaling from 6K to 50K TPS via rSocket migration supporting ~1.5M concurrent connections at sub-25ms TP99. The challenge of maintaining product quality and user trust at that scale, while shipping rapidly, is directly transferable to what Inflection is building. **Selected Prior Experience** - **Multi-agent orchestration at production scale:** Built OpenClaw multi-agent orchestration framework with gateway protocol, subagent delegation, profile management, and session switching — enabling coordinated AI agent workflows across real estate, insurance, health/dental, and financial markets verticals. - **LLM product from 0-to-1:** 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 for Fintellect AI — then took it through App Store launch. - **Consumer mobile and web at scale:** Delivered cross-platform features across ~20 mobile apps and 30+ product SKUs at Intuit; implemented Background-to-Foreground Messaging on iOS/Android with sub-100ms latency; shipped ICE Presence in async chat generating $480K/month in additional invoicing. - **Developer platform and SDK leadership:** Extended Java and Python SDK Starter Kits with scaffolding templates, build configurations, testing frameworks, and CI/CD integration — reducing developer onboarding from 2–3 weeks to minutes. Delivered ICE Self-Service platform (DevPortal, GitOps config, ICE Playground), mitigating $1M+ in projected opex growth. - **Data-driven product decisions:** Used SQL and BigQuery to prioritize developer pain points across platform infrastructure; designed RICE-based prioritization frameworks for multi-service backlogs at Splunk, balancing internal partner, third-party developer, and Fortune 500 customer requirements. - **Fine-tuning and open-source model fluency:** Built hands-on experience with TRL, VeRL, OpenRLHF, and NeMo RL — the same open-source fine-tuning ecosystem (Nemotron, Qwen) referenced in Inflection's preferred qualifications — with direct benchmarking experience across proprietary vs. open-source trade-offs. - **A/B testing and experimentation infrastructure:** Built statistical rigor into aeval with bootstrap confidence intervals, effect size measurement, saturation detection, and regression detection — the same discipline required to run large-scale experiments on Pi's user experience. **Closing** Inflection's mission — combining EQ and IQ to elevate human potential — isn't a product positioning statement. It's a hard technical and design problem that requires someone who can sit in an architecture review on constrained inference, write a crisp PRD for a memory and profile system, and still ask the right question about whether the feature actually *feels* right to the person using it at 11pm when they're working through a difficult decision. That's the work I want to do, and I believe my background positions me to contribute meaningfully from day one. I would welcome the opportunity to discuss how my experience maps to Inflection's roadmap in more depth. Sincerely, **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info