cover_letter / art_keKGOyC2TFg

Dear Faire Hiring Team, Faire is doing something genuinely hard: bringing structure, discovery, and data intelligence to a wholesale market that has operated on relationships and gut instinct for generations. The mission — leveling the playing field so independent retailers can compete and local communities can thrive — is one I find compelling precisely because the technical challenge is non-trivial. My path from hand-coding backpropagation through time in C++ at UC Berkeley in 2004 to building production RL post-training workbenches and AI evaluation platforms today has been defined by exactly this kind of problem: using machine learning to bring order and signal to noisy, high-dimensional spaces where the stakes for real people are high. ## Technical and ML Foundation My ML work is grounded in systems I have built end-to-end, not integrated. My NeurIPS 2014 paper on artificial neural networks for protein secondary structure prediction — work that began with a hand-coded neural network in C++ with custom BPTT at Berkeley — established a foundation in model architecture, evaluation rigor, and the gap between benchmark performance and real-world utility that has shaped every product decision I have made since. The 2026 rewrite of that system spans 413 parameters to 8 billion (a 19-million-fold scale increase), implemented in PyTorch with five neural architectures, MLflow experiment tracking, Optuna hyperparameter optimization, and FastAPI serving — 823 automated tests across six Docker containers. More directly relevant to Faire's search context: 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 including bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, and saturation detection. The CI/CD integration includes regression detection and automated safety gates — the kind of offline metric infrastructure that is foundational to iterating on search ranking systems responsibly. I also built an **RL post-training workbench** implementing 12 algorithms (PPO, GRPO, DAPO, REINFORCE, REINFORCE++, RLOO, DPO, SimPO, IPO, KTO, ORPO, SPPO) with standardized throughput, memory, and convergence benchmarking across TRL, VeRL, OpenRLHF, and NeMo RL — work that required me to deeply understand reward modeling, policy optimization tradeoffs, and the relationship between offline metrics and online behavior. ## Why This Role The arc from building ML evaluation infrastructure to owning search algorithm product strategy at Faire is a direct one. At Intuit, I spent three years as a Staff PM on developer platform infrastructure, where I learned that the difference between a search or discovery system that works and one that scales is almost entirely in how well you instrument it, segment query intent, and build feedback loops between offline evaluation and online experimentation. Faire's search problem — connecting retailers to the right brands and products across a fragmented, long-tail catalog — is exactly the kind of high-dimensional retrieval and ranking challenge where that discipline matters most. What specifically excites me about this role is the combination of query segmentation strategy and metric refinement. The JD calls out diagnosing root causes for today's failure modes and refining both offline and online metrics — this is the work I find most intellectually engaging, because it requires holding the user's mental model (a retailer searching for "coastal home goods" or "sustainable kids apparel") alongside the system's representation of that intent, and closing the gap between them systematically. The opportunity to work directly with founders and the CEO on decisions of this scope, at a company with Faire's market position, is rare. ## Selected Relevant Experience - **Intuit — ICE Platform:** Achieved 275% YoY growth in ICE engagements, scaling to 675M+ in FY23 across QuickBooks, TurboTax, Mint, Mailchimp, and Credit Karma; scaled throughput from 6K to 50K TPS via rSocket migration supporting ~1.5M concurrent connections with sub-25ms TP99. Built and maintained telemetry pipelines (SQL, BigQuery) to prioritize developer pain points across ~20 mobile apps and 30+ product SKUs. - **Splunk — Search Orchestration:** Owned Search Service (Go microservices), Search Catalog (PostgreSQL metadata service), and Splunk Processing Language (SPL/SPL2). Led query performance optimization initiative achieving up to 10x performance improvements in Splunk Cloud Services search. Delivered Scheduler Service end-to-end in ~4 months. Designed repeatable RICE-based prioritization framework for three microservice backlogs. - **aeval — Evaluation Platform:** Built offline evaluation infrastructure with statistical rigor (bootstrap CIs, Welch's t-test, Cohen's d), adversarial safety testing, and CI/CD regression detection — directly analogous to the offline metric suite Faire needs to understand search component performance before shipping. - **RL Workbench:** Implemented 12 RL algorithms with cross-tab workflow lineage tracking and standardized benchmarking across four frameworks — demonstrates the ability to reason about algorithm tradeoffs (velocity, convergence, memory) at the level this role requires when evaluating ranking model approaches. - **Fintellect AI — RAG Pipeline:** 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 — directly relevant to the retrieval and ranking infrastructure underlying modern search systems. - **Intuit — Developer Assessment:** Conducted enterprise-wide Service Language Assessment across 9 languages, analyzing usage data and developer feedback to inform strategic investment decisions presented to the CTO — evidence of the structured analytical approach I bring to complex, ambiguous prioritization problems. - **Kaiser Permanente — Platform Scale:** Led Splunk Logging-as-a-Service at 1.7 TB daily volume across 200+ internal enterprise customers, including demand forecasting and capacity planning across multiple datacenters — grounding in the operational realities of high-scale data systems. ## Closing Faire's bet is that technology can do for independent retail what it has done for every other fragmented market: create transparency, surface signal, and let the best products win regardless of who has the largest sales force. Search is the mechanism through which that bet is realized or lost for every retailer who opens the platform. I want to own that mechanism — to build the query segmentation frameworks, the metric infrastructure, and the algorithm strategy that make discovery genuinely work at Faire's scale. I would welcome the opportunity to discuss how my background maps to the specific challenges your team is navigating. Thank you for your consideration. --- **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info