• Designed and built an internal AI sales assistant (EB-Copilot) using a FastAPI (Python) backend with a provider-agnostic LLM orchestration layer, enabling the sales team to query customer data, credit limits, and purchase history in natural language — embedded into the company portal via iframe with Server-Sent Events (SSE) for real-time streaming responses.
• Implemented bank file-based integrations (CNAB format) for automated payment remittance, payment collection, and bank reconciliation — eliminating manual statement processing and enabling automated matching of bank transactions against internal financial records.
• Architected and shipped an offline-first Flutter mobile application for remote field operations, implementing local-first data persistence and automated background synchronization with enterprise ERP systems — enabling reliable data collection in areas with no connectivity.
• Engineered backend integrations with the company’s core management platform and Senior ERP, developing custom business rules, data flows, and secure API endpoints to bridge operational systems.
• Built an internal ticketing system with dynamic department-based SLA tracking and performance dashboards, improving accountability and operational visibility across support teams.

• Developed a .NET automation tool for mass video overlay generation, reducing production processing time by 87.5% during peak digital catalog periods.
• Built custom integrations to extract data securely from the auction management system and feed it directly into live broadcast graphics — eliminating manual data entry in a time-critical, zero-error-tolerance environment.
• Regularly identified workflow gaps in live production and delivered custom tooling to close them, operating in time-critical live broadcast environments.

A database-free system that composites a custom overlay image onto a video using FFmpeg. Users upload files directly to S3 via presigned POST URLs; the pipeline handles rendering fully asynchronously with no persistent infrastructure beyond object storage.

• Architecture: Event-driven pipeline — S3 Event Notifications → SQS → Fargate Worker. Job state is inferred from object storage rather than persisted in a database.
• Backend: .NET 10 Minimal API on AWS Lambda (presigned URL generation, read-time status inference); .NET 10 Worker on AWS Fargate / Docker (SQS consumer, S3 distributed locking, idempotency guard, FFmpeg execution).
• Frontend: Astro (static shell) + React Island (interactive canvas designer for overlay creation).
• Infrastructure: Pulumi IaC with isolated dev and production stacks.
• Quality: Outside-in TDD strategy (Acceptance → Integration → Unit) using WebApplicationFactory and LocalStack for cloud emulation. Documented major architectural trade-offs using Architecture Decision Records (ADRs).