InfiniBandLed the design and implementation of InfiniBand for Crusoe’s GPU clusters, achieving multi-terabit interconnection speeds. This critical infrastructure supports large-scale AI training on cutting-edge Nvidia hardware and plays a pivotal role in attracting and retaining top AI clients. Data WarehouseBuilt Crusoe’s data warehouse, establishing it as the authoritative source for cloud usage and revenue data. The system processes relational and time series data, providing comprehensive reports and dashboards to teams across the company.ObservabilityLed an initiative to modernize, standardize and consolidate observability time series data on a modern, Prometheus-compatible stack. The project enabled the company to set and measure adherence to SLA targets.

Smart ProcessingArchitect of strategic company initiative to automate complex business processes. Defined overall architecture, involving changes to a dozen applications, for this high visibility, multi-year program involving over 100 engineers. Collaborated closely with business partners on the optimal approach to significantly reduce annual run rate, and scale to handle higher workloads with minimal additional headcount. Oversaw application architects working on detailed design changes to ensure adherence to project and enterprise architecture.

API GatewayArchitect of the firm’s new API Gateway infrastructure. The API Gateway allows clients to authorize third parties to access their financial data in a secure and transparent manner. Design took a defense in depth approach, with multiple layers of protection across the system. Led the API Gateway vendor selection and contract negotiations. Defined the deployment topology to meet the 99.99% availability target. Voting member of FDX (Financial Data Exchange), an industry group defining common APIs for safe data sharing between financial institutions and FinTech. The project reduced a significant risk exposure to clients and the firm, and established an enterprise gateway capability. Options Spreads in IRAsArchitect for a multi-year, multi-million dollar project, involving two dozen applications, to add much requested options spread trading capabilities to IRAs. The project navigated complex legal and regulatory challenges with sometimes conflicting rules. The implementation monitored client risk in near real-time. Identified risks were handled by a combination of automated and high-touch manual client outreach. Risk reduction trades were made if necessary - a particularly useful feature in fast moving markets. Calculating near real-time risk assessments at scale was a significant engineering challenge necessitating sophisticated caching and distributed calculation engines. Real-time Cost BasisArchitect of a new real-time cost basis calculation engine for Schwab clients. Prior to this project, cost basis was only updated daily for most accounts. The implementation cached significant quantities of data on a new, ultra-fast caching infrastructure and scaled out calculations across a highly distributed compute cluster. The project resolved a long-standing client pain point and helped attract and retain clients.

Fixed Income Trading platform – New IssuesArchitect of Schwab’s Fixed Income trading platform. Collaborated with industry partners to offer New Issue Municipal securities at Schwab. This is the municipal bond equivalent of a company IPO, and the innovation propelled Schwab to become the industry leader offering online trading of New Issues directly to clients and advisors. Implementation required new synchronous and asynchronous connectivity with third parties, new workflow capabilities and changes to several internal applications across the firm. Mutual Funds BillingLead developer/architect of a new Mutual Fund Billing system to invoice fund companies. The requirements were somewhat unique in that the system had to be capable of recreating an invoice months later, using the data and invoicing rules known to the system at that time, or with updated data or rules. This ability enabled the business to easily recover if data from an upstream system was later corrected, or an invoicing rule required adjustment. The invoicing system processed over a billion rows of data nightly and executed in parallel cross several nodes, with automated recovery from node failure.