A Netty-based middleware proxy that duplicates live production traffic to shadow test databases, so schema migrations and chaos testing can be rehearsed against real query patterns without touching production.

• 3ms average latency overhead at 50K+ requests/sec, using Netty's asynchronous non-blocking I/O for interception and forking.
• Real-time Prometheus and Grafana dashboards comparing production against shadow query performance, surfacing 12+ regressions before they reached production.

• Java
• Netty
• Kafka
• PostgreSQL
• MongoDB
• Prometheus
• Grafana

An AML/OFAC compliance agent that detects Bank Secrecy Act structuring patterns — deposits deliberately kept under reporting thresholds — and produces reports a compliance officer can actually file.

• 94% precision on suspicious-transaction detection, by orchestrating multi-step LLM reasoning and tool-calling through LangGraph rather than a single prompt.
• Every verdict carries its evidence chain, so the output is auditable instead of a bare score.

• Python
• LangGraph
• AWS Bedrock
• FastAPI

A natural-language workflow engine that turns plain-text process descriptions into validated, exportable swimlane blueprints — aimed at business analysts who know the process but not the tooling.

• Cut workflow design turnaround from hours to under 5 minutes by removing manual Step Functions configuration entirely.
• Ambiguous descriptions are recompiled from a clarified source rather than hand-patched, so the blueprint always matches the text that produced it.

• JavaScript
• NLP
• Workflow engines

PageRank implemented over large synthetic web graphs, built to evaluate distributed graph-processing frameworks for research teams rather than to serve traffic.

• Achieved linear scalability across node counts by tuning Spark partition strategy and shuffle behaviour on EMR clusters.
• Shuffle cost, not compute, turned out to be the binding constraint — the partitioning work is where the speedup lives.

• Apache Spark
• AWS EMR
• S3