cover_letter / art_a99MJ7qLf8w

Dear Dealpath Hiring Team, Dealpath sits at a genuinely interesting intersection: the world's largest asset class, a platform trusted by Blackstone and Nuveen, and a moment where AI can meaningfully compress the time between data and decision. That combination is what drew me to this role. My path from hand-coding backpropagation in C++ at UC Berkeley in 2004 to building production multi-agent orchestration systems and RL post-training workbenches today has been defined by exactly this kind of inflection point — where deep technical work meets high-stakes domain problems. --- ## Technical and AI Foundation My AI/ML work spans research and production. In 2014, I published at NeurIPS on artificial neural networks for protein secondary structure prediction — work that began with a hand-coded BPTT implementation in C++ and was rewritten in 2026 as a full PyTorch platform 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 recently, 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 baked in — bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, and automated regression gates in CI/CD. This is precisely the kind of eval infrastructure Dealpath's JD references: building and iterating on AI evals, diagnosing model failures, and driving continuous improvement before features reach production. I also built an **RL post-training 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 for real TRL-powered GRPO/DPO training with live SSE metric streaming on Apple Silicon and CUDA; and an Arena for head-to-head framework benchmarking across TRL, VeRL, OpenRLHF, and NeMo RL with GPU passthrough in Docker. Twelve RL algorithms implemented, with standardized throughput, memory, and convergence benchmarking across frameworks. On the agentic side, I designed and shipped **OpenClaw**, a multi-agent orchestration framework with a gateway protocol, subagent delegation, profile management, and session switching — enabling coordinated AI agent workflows across real estate, insurance, health/dental, and financial markets. I also architected a 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. --- ## Why This Role Dealpath's challenge is not generic AI adoption — it is applying AI to a domain where data is proprietary, workflows are sophisticated, and the clients (Blackstone, Nuveen, Starwood) have zero tolerance for hallucination or miscalibrated outputs. That requires exactly the combination I have built toward: deep enough technical fluency to participate credibly in systems and agent design, enough product discipline to know when to automate and when to keep humans in the loop, and enough domain proximity to real estate to move quickly. I hold a California Real Estate Broker/Sales Agent license and have built production real estate tooling — including automated CMA report generation via Redfin and Zillow APIs and AI-powered real estate agents within StreamIO — so the deal lifecycle from sourcing through asset management is not abstract to me. The opportunity to apply AI to underwriting workflows, deal screening, and portfolio monitoring at the scale Dealpath operates is a genuinely compelling product problem. --- ## Role-Specific Connection Three things in the JD stand out specifically. First, the emphasis on building and maintaining AI evals to validate model performance — this is infrastructure I have built from scratch and understand as a first-class product concern, not an afterthought. Second, the expectation that the PM will prototype and vibe-code proof-of-concepts before committing to a full build — I ship production code across TypeScript, Python, Go, and Swift, and have done exactly this in both StreamIO and Fintellect AI. Third, the external thought leadership dimension — I have spoken at DeveloperWeek 2022 and Splunk .conf18 and .conf19, and I teach cloud computing, data analytics, and Java programming as adjunct faculty at De Anza College, so translating complex technical concepts for diverse audiences is a practiced skill, not an aspiration. --- ## Selected Prior Experience - **OpenClaw multi-agent orchestration** (StreamIO AI): Designed gateway protocol, subagent delegation, profile management, and session switching — enabling coordinated AI agent workflows across real estate, insurance, and financial markets industries. - **RAG pipeline + LLM orchestration** (Fintellect AI): Architected ChromaDB vector store retrieval, multi-provider LLM orchestration (Claude, GPT-4, Gemini) with fallback routing, structured output validation, and token budget optimization for a production financial advisory platform. - **aeval — AI Model Evaluation Platform**: Built local-first eval platform with adversarial safety testing, refusal detection, data contamination detection via SHA-256 hashing, and CI/CD integration with automated regression gates. Stack: FastAPI, TimescaleDB, Redis, Next.js, Ollama. - **Real estate AI tooling** (StreamIO AI): Built automated CMA report generation via Redfin/Zillow APIs and AI-powered real estate agents with contextual conversation and document analysis — directly applicable to Dealpath's deal lifecycle workflows. - **ICE Self-Service platform** (Intuit): Delivered developer platform reducing onboarding from 2–3 weeks to minutes in pre-prod; scaled to 675M+ engagements in FY23 across QuickBooks, TurboTax, Mint, Mailchimp, and Credit Karma — demonstrating ability to ship platform infrastructure at enterprise scale. - **Scheduler Service delivery** (Splunk): Delivered end-to-end in ~4 months, enabling scheduled search capabilities for first-party applications; led query performance optimization achieving up to 10x improvements for a Fortune 500 beta customer. - **NeurIPS 2014 publication**: Accepted paper on artificial neural networks for protein secondary structure prediction — establishing a research foundation that informs how I think about model quality, evaluation, and the gap between benchmark performance and production reliability. --- ## Closing Dealpath's mission — uniting data, insights, and execution for the world's largest asset class — is the kind of problem that benefits from a PM who can hold both the technical depth and the domain context simultaneously. I have spent the last two years building AI systems that are close to the work: writing evals, debugging agent failures, shipping RAG pipelines, and iterating with real users. I would welcome the opportunity to bring that to Dealpath's AI roadmap. Thank you for your consideration. --- **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info