Technical Demonstrations & Selected Work.

Problem Statement: Under simulated concurrent traffic of 2,500 requests per second, a standard Entity Framework Core backend experienced severe SQL connection pool starvation, elevated memory allocations, and p95 latency exceeding 320ms.

Solution Overview: Re-architected the read/write pathways using Clean Architecture and the CQRS pattern via MediatR. Implemented Dapper micro-ORM for read projections, added Redis distributed cache-aside for high-frequency endpoints, and optimized SQL Server connection pooling and query parameterization.

Problem Statement: A high-volume transactional database table with over 4.5 million records was locking during daily reporting queries, causing production timeouts, 95% CPU spikes, and severe lock escalations.

Solution Overview: Identified missing composite indexes using Extended Events and Query Store. Re-engineered correlated subqueries into Common Table Expressions (CTEs), introduced filtered indexes on status flags, and updated index statistics with a targeted maintenance job.

Problem Statement: Consolidating disparate daily transaction dumps (REST APIs, CSV uploads, legacy database replicas) required hours of fragile manual data cleaning, resulting in silent data drift and inaccurate reporting.

Solution Overview: Architected an idempotent Python-based data ingestion pipeline. Built with Pydantic for strict schema enforcement, automated dead-letter queues for corrupted records, and automated loading into analytical warehouse schemas with complete audit logging.

Problem Statement: Standard LLM implementations frequently hallucinate operational details and lack trustworthy references when searching through large corporate policy, compliance, or API documentation.

Solution Overview: Engineered a hybrid retrieval pipeline combining dense vector embeddings (cosine similarity) with sparse BM25 keyword matching. Grounded LLM responses strictly inside retrieved chunks with exact document and section citations.

Problem Statement: Enterprise application tied to Windows Server with legacy WCF services, blocking containerization and incurring high Windows Server licensing and VM maintenance overhead.

Solution Overview: Applied the Strangler Fig pattern to migrate legacy WCF services into gRPC and modern REST endpoints on .NET 8. Ported business libraries to .NET Standard 2.0 / .NET 8, containerized with Alpine Linux Docker images, and deployed to lightweight cloud runners.

Problem Statement: Accounts payable staff spending 20+ hours per week manually extracting vendor details, line items, tax numbers, and payment terms from scanned and digital PDF invoices.

Solution Overview: Developed an automated document pipeline that ingests multi-format invoices, applies layout-aware OCR and structured prompt schema extraction, validates calculations, and queues verified entries directly for ERP entry.