MLOps engineer · reliability, governance, operations

I turn model experiments into governed systems.

I’m Sushan Manandhar. I build reproducible ML delivery paths with point-in-time data, measurable release gates, observable services, and recovery controls—then publish the artifacts that verify them.

Review engineering evidence →Resume ↓LinkedIn ↗ Contact
222 automated tests3 core deep dives5 checksum-verified downloads
Operational lifecycle

Evidence across the full loop.

Three projects demonstrate governed delivery, evaluation-first LLMOps, and risk-controlled paper operations. ModelGuard remains a focused supporting security artifact.

01ValidateSchema, policy, and safety checks
02TrainReproducible data and experiments
03ReleaseQuality and regression gates
04ServeTyped APIs and containers
05OperateMetrics, drift, audit, response
Selected work

Three evidence-first deep dives.

View full case studies →
End-to-end MLOps

FraudShield

Reproducible fraud-risk training, promotion, serving, monitoring, autoscaling, and infrastructure assets.

MLflowDVCKubernetesTerraform
0.838 controlled ROC-AUC · 4 tests
LLMOps

RAGOps

Source-grounded retrieval with citations, abstention, safety checks, evaluation, and release gates.

RAGFastAPIPrometheus
7 evaluation cases · 6 tests
Operations

SignalOps

Authorized alert ingestion, deterministic paper simulation, risk execution stress, exposure controls, gap-aware exits, kill switches, and restart-safe reconciliation.

Discord BotSQLitePrometheus
Independently verifiable evidence · 206 tests