Setiawan Muhammad

Audited and validated large batches of LLM-generated code implementations against official language/library specifications and documentation, identifying subtle API hallucinations, concurrency bugs, and edge-case failures. Authored deterministic unit and integration tests to evaluate model outputs on complex algorithms and system design tasks, delivering verified ground-truth datasets for frontier model alignment benchmarking.

Investigated underlying causes of production bottlenecks and infrastructure incidents using root-cause troubleshooting rather than temporary patches, including mitigating rogue flood packet behavior and resolving UI rendering lag issues through profiling. Built fleet-wide API observability and telemetry using high-throughput ingestion and WebSocket streaming to support long-running uptime. Conducted AI red teaming with adversarial evaluation of LLM-integrated services and implemented guardrails to restrict unauthorized tool execution across autonomous workflows. Maintained zero-downtime blue/green deployment pipelines via CI/CD with automated unit tests, secret scanning, and reliability-focused Git workflows.

Operated as a teaching assistant and lab mentor for multiple courses, handling code reviews for hundreds of student repository submissions per semester. Guided peers on Git hygiene, algorithmic efficiency, and systematic root-cause debugging. Authored lab instructions and troubleshooting guides covering race conditions, memory leaks, and packet analysis practices, improving practical pass rates for students.