Wyatt Walsh

Sole AI/ML Engineer delivering production applied AI for risk/compliance and growth at a high-risk-merchant payments platform. Ship document intelligence and restricted-item compliance screening for merchant onboarding using agentic LLM workflows and multimodal inputs, with human-in-the-loop UX and audit-aware handling of PII and confidential documents.

• Deployed submission-time document-category validation using PydanticAI, Amazon Bedrock AgentCore, and Claude, surfaced as in-browser feedback for confidential uploads; reduced analyst follow-ups and resubmission requests ~90% by blocking miscategorized documents before they entered review queues.
• Built restricted-item compliance screening that combines product images with structured submission metadata; translated banking-partner policy requirements into automated checks supporting onboarding across tens of merchants, multi-location operations, and catalogs with hundreds of items per location.
• Built a NotebookLM-based technical knowledge hub spanning vendor/web documentation and internal monorepo docs; automated corpus refresh with a Colab pipeline that re-ingests the latest sources to keep retrieval accurate and up to date.
• Drove internal AI adoption with practical enablement assets: Notion playbooks covering MCP onboarding, “best MCPs” recommendations, and high-signal prompt patterns; ran weekly office hours reaching ~15% of the company and standardized PII/confidential-document safeguards and auditability requirements.

Engineered enterprise-grade data pipelines, data integrations, and analytics solutions for trading, quantitative research, and risk groups within the credit division of the investment bank and fostered improved AI development tool usage across the firm.

• Pioneered firm-wide rollout of generative AI development tools (LLMs, GitHub Copilot, Codeium) as one of the first 500 engineers, delivering feedback that shaped secure, scalable deployment and leading internal workshops to boost productivity for technical and non-technical teams.
• Devised ML-driven anomaly detection framework for data health monitoring, reducing false positives by 25% and cutting manual review time, empowering risk teams to focus on critical alerts.
• Developed resilient Python wrapper for KDB+, democratizing data access for 5,000+ users with CI-style tests and optimized Q functions, reducing manual processing by over 10 hours per week.
• Engineered a cloud-function–driven asset-fee data sink, consolidating 10+ sources into a centralized database, cutting manual maintenance by 90% and providing on-demand access for traders through a Python SDK, REST API, and interactive dashboard.

Delivered an advanced DeFi DApp as the founding engineer — integrating ethers.js for gas-optimized web3 smart contract interactions, implementing CI/CD and TDD pipelines to boost development efficiency, and partnering with founders to translate Figma designs into pixel-perfect interfaces.

• Architected a scalable NFT-collateralized lending DApp using React, Next.js, TypeScript, MobX, and ethers.js, enabling secure, performant on-chain loans.
• Implemented end-to-end CI/CD with GitHub Actions, Husky, Cypress, and Jest to halve release cycles and raise test coverage to 90%, while mentoring a front-end intern to build key UI components.
• Optimized front-end performance through code splitting, lazy loading, and asset caching, reducing initial load times by 40% and JavaScript bundle size by 30%.

Enabled better data-driven decision making for Web3 stakeholders by tokenizing a 1M+ post/comment, Web3-related Reddit dataset onto a decentralized data marketplace after creating a public-facing web app and securing grant funding to enable the project.

• Engineered and launched a Next.js/TailwindCSS marketing site with top-tier Lighthouse scores and secured $75K in OCEAN tokens via a competitive DAO grant.
• Engineered Python pipelines to extract and normalize a 1M+ Web3-related Reddit posts & comments dataset and published the data on Ocean Protocol’s Data Marketplace.

Engineered predictive demand forecasting models, improving forecast accuracy by 5–25% to optimize inventory planning and reduce stockouts across multiple retail divisions.

• Applied time-series decomposition and bias correction techniques to demand forecasts, boosting accuracy by up to 25% and minimizing surplus inventory costs.
• Architected scalable data infrastructure using MySQL and Apache Spark to process 5+ years of sales data, improving data processing throughput by 95% and standardizing insights across forecasting teams.