cover_letter / art_e-n8Z3V1_nA

Dear Pinterest Hiring Team, Pinterest sits at a rare intersection — a platform where users arrive with genuine intent, not passive consumption, and where the journey from inspiration to purchase spans sessions, moods, and months. That statefulness problem — helping someone pick up exactly where they left off, refine a search, broaden it, and ultimately act — is one of the most technically and product-strategically interesting challenges in consumer AI today. My interest in this role is grounded in a specific arc: I spent three years at Intuit scaling a developer infrastructure platform to 675M+ engagements, learning how personalization signals compound at scale, and I have spent the past year building AI products end-to-end — from retrieval pipelines to RL post-training workbenches — that directly inform how I think about relevance, ranking, and user intent. **AI/ML and Technical Foundation** My technical credibility in AI/ML spans from first principles to production systems. In 2004, I hand-coded a neural network in C++ with custom backpropagation through time for protein structure prediction — work that was accepted at NeurIPS 2014. In 2025–2026, I rewrote that system in PyTorch, scaling from 413 to 8 billion parameters across five architectures (feedforward, GRU, Transformer, ESM-2, multi-task), with MLflow experiment tracking, Optuna hyperparameter optimization, and FastAPI serving across six Docker containers. More directly relevant to search personalization: I built aeval, a local-first AI model evaluation platform with five core eval types — factuality, reasoning, instruction-following, safety, and code generation — with statistical rigor including bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, and saturation detection. This is the kind of measurement infrastructure that underpins any serious relevance improvement program. I also built an RL post-training workbench that benchmarks GRPO and DPO across TRL, VeRL, OpenRLHF, and NeMo RL, implementing 12 RL algorithms (PPO, GRPO, DAPO, REINFORCE, DPO, SimPO, IPO, KTO, and others) with standardized throughput, memory, and convergence benchmarking. Understanding how reward functions shape model behavior is directly applicable to how ranking models learn from implicit user signals like saves, clicks, and session depth. For Fintellect, I architected a RAG retrieval pipeline with ChromaDB vector store, multi-provider LLM orchestration across Claude, GPT-4, and Gemini, fallback routing, structured-output validation, and token-budget optimization — shipping 13 specialized AI advisors with context-aware, guided advisory flows. Building retrieval systems that surface the right content for a user's specific intent, not just their query string, is precisely the problem Pinterest Search Personalization is solving. **Bridge** The through-line in my career is building AI-powered systems that close the gap between what a user is trying to accomplish and what a platform surfaces for them — whether that is a developer finding the right SDK, a trader finding the right instrument, or a Pinner finding the product that matches the aesthetic they have been curating across boards for six months. This role is the natural next chapter of that work, at a platform where the personalization signal is uniquely rich and the stakes for getting relevance right are directly tied to user joy and commercial outcomes. **Why This Role at Pinterest** What excites me specifically about the Search Personalization mandate is Pinterest's statefulness advantage. The ability to model multi-session journeys — where a user narrows from "living room ideas" to "mid-century modern sofa under $1,200" across weeks — is a personalization problem that most platforms cannot even frame correctly because they lack the save/board/collaboration graph that Pinterest has. Incorporating that long-horizon intent signal into search ranking models, and building the product instrumentation to measure whether relevance improvements translate to downstream shopping actions, is exactly the kind of 0-to-1 product strategy work I have done repeatedly. I am also drawn to the cross-functional scope: partnering with data science on model features, with design on query refinement UX, and with engineering on the ML infrastructure that makes real-time personalization possible at Pinterest's scale. **Selected Prior Experience** - **Intuit — 675M+ engagements, 275% YoY growth:** Scaled the ICE platform to 675M+ engagements in FY23 across QuickBooks, TurboTax, Mint, Mailchimp, and Credit Karma; drove 275% YoY growth by identifying and resolving developer pain points through telemetry and SQL/BigQuery usage analysis across ~20 mobile apps and 30+ product SKUs — directly analogous to using behavioral data to prioritize search relevance improvements. - **Intuit — throughput scaling:** Scaled platform throughput from 6K to 50K TPS via rSocket migration supporting ~1.5M concurrent connections with sub-25ms TP99 — experience with the infrastructure constraints that bound real-time personalization systems. - **aeval — model evaluation platform:** Built statistical evaluation infrastructure (bootstrap CIs, Welch's t-test, Cohen's d, saturation detection) with CI/CD regression detection and automated safety gates — the measurement discipline required to run rigorous search relevance experiments. - **RL Workbench — reward function design:** Built Reward Lab for designing and A/B testing reward functions (RLVR, learned, hybrid) across GSM8K, MATH, HumanEval, and UltraFeedback datasets — translatable to designing reward signals from implicit Pinner feedback (saves, closeups, outbound clicks, purchases). - **Fintellect — RAG retrieval and multi-provider LLM orchestration:** Architected ChromaDB-backed retrieval pipeline with structured-output validation and context-aware advisory flows — directly relevant to query understanding and result relevance in a personalized search context. - **Splunk — query performance and search infrastructure:** Owned Search Service (Go microservices), Search Catalog (PostgreSQL metadata service), and SPL/SPL2; led query performance optimization achieving up to 10x improvements for a Fortune 500 beta customer — foundational search infrastructure experience. - **Splunk — RICE prioritization across microservice backlogs:** Designed repeatable RICE-based prioritization framework balancing internal partner, third-party developer, and Fortune 500 customer requirements across three microservice backlogs — the structured decision-making approach this role requires when trading off personalization model investments. **Closing** Pinterest's mission — bringing everyone the inspiration to create a life they love — is one that takes search personalization seriously as a product discipline, not just a ranking problem. The users who come to Pinterest with active intent deserve a search experience that understands not just their query but their journey, their taste, and where they are in the decision process. I have spent 12 years building the technical foundation and product judgment to contribute meaningfully to that mission, and I would welcome the opportunity to bring that experience to the Search Personalization team. Thank you for your consideration. --- **O. Felix Amoruwa** famoruwa@berkeley.edu · 909-731-9011 · felixamoruwa.info