Resume

• Built process health monitoring for a $600M+ file exchange network. 300+ daily regulatory transmissions had no unified failure tracking. Built a pipeline-backed anomaly detection system with predictive validation. Reduced manual monitoring effort by 78%.
• Resolved a 20,000+ account logic failure. Data team was backlogged 6+ weeks. Built SQL framework to isolate root cause, remediate impacted accounts, and document failure taxonomy. Process team now owns root-cause analysis.
• Validated specifications for the $35B+ Discover merger integration. Owned technical validation of the Systems Integration Guide for legacy firm onboarding. Surfaced design-layer gaps in regulatory file exchange formats with Engineering and vendor partners. Zero transmission errors through cutover.
• Designed a 5-month modernization roadmap for fragmented manual workflows. Staged automation with dependency sequencing, stakeholder gates, and rollback protocols. Learned Splunk, Postgres, and JSON parsing to unblock 6 months of stalled monitoring work.

• Promoted four times in five years -- from teller through loan underwriting and account management, to full branch operational strategy, then enterprise-level operational support across a 25-branch network during a major merger.
• Completed B.S. at George Mason University concurrently while carrying a full-time operational role across all four promotion stages.
• Built compliance tracking and reporting dashboards (Power BI) for regulatory documentation (NCUA, CFPB). Designed monitoring infrastructure and workflow automation that reduced complaint response cycles by 50%.
• Proposed branch-wide modernization roadmap with VP and SVP backing. Designed automation strategy for manual workflows across the branch network. Approved and moved into implementation.

Open-source desktop application (Electron + React + Node.js) integrating chat, terminal, notes, and persistent memory into a single local-first interface.

• Built a reframe detection engine that classifies when language models subtly rewrite user statements. Triggers automatic prompt correction when user-rated inaccuracy exceeds 40%.
• Designed provider-agnostic model routing across Anthropic, OpenAI, and Ollama. Auto-selects by intent complexity and cost. Runs air-gapped on local models or through cloud APIs.
• Designed hierarchical memory compression -- raw to summary to pattern to principle. Strength decay with retrieval reinforcement for long-term context retention.
• Built post-session extraction pipeline. Extracts decisions, tasks, and code artifacts on conversation end. Generates typed notes and indexes into FTS5 for search.

Open-source local infrastructure for memory, routing, and provenance. Monorepo architecture -- MIT-licensed memory primitive, AGPL core.

• Built the MCP Memory Server (MIT). SQLite-backed persistent memory implementing the Model Context Protocol. Strength decay, retrieval reinforcement, and semantic search.
• Designed dual integration architecture -- MCP for ecosystem compatibility plus a localhost proxy for conversation intercept and PII scrubbing.

Hand-built with Astro and vanilla JS. WebGL2 ASCII shader over a Gray-Scott reaction-diffusion field in a Web Worker. Audio-reactive via Web Audio API.