- Verified in testsPipeline and release gate
4 tests; DVC and MLflow tracking, regression policy, and a containerized FastAPI package.
- Current boundarySynthetic evaluation
Held-out synthetic metrics only. Infrastructure and monitoring are checked definitions, not a live deployment.
- Next proofOperational staging
Validate the service and telemetry in a real staging environment with data rights and recovery drills.
My MLOps skills, with evidence for each one.
My stack spans data validation, experimentation, release automation, and operating controls. I link each area to a project and label the evidence clearly: tested code, checked infrastructure definitions, or foundational experience.
Built, bounded, next.
Three core projects, each shown with its tested evidence, present limit, and next validation step. The labels describe evidence, not a percentage of completion.
- Verified in testsEvaluation and refusal gate
6 tests and 7/7 synthetic evaluation cases; citation, retrieval, safety, and latency checks.
- Current boundarySmall controlled corpus
The demo uses deterministic retrieval and app-level filters, without a hosted LLM or field evaluation.
- Next proofRepresentative evaluation
Expand the corpus and test model-backed answers against harder, independently reviewed cases.
- Verified in testsControls and data contracts
460 tests; deterministic risk, simulated fills, provenance, and offline crypto identity checks.
- Current boundaryNo market validation
Crypto fixtures are synthetic. No live collection, return, predictive score, or deployed AWS stack is claimed.
- Next proofPoint-in-time data
Admit authorized historical observations, bind DEX pairs to verified identities, then evaluate unseen outcomes.
Supporting security artifact: ModelGuard case study has 6 passing tests for fail-closed model-repository admission. It is not a fourth core deep dive.
Where I applied each skill.
Status labels show which systems I built and tested. Cloud definitions have passed code checks, but the AWS services have not been deployed.
I package services with Docker and define Kubernetes workloads. In SignalOps, I test queue retries, dead-letter handling, and a paper-only execution boundary.
Evidence: SignalOps operations layer →I use purged walk-forward evaluation, calibrated baselines, and signed reports. I bind each candidate to its inputs and protect a final holdout from repeated tuning.
Evidence: research lifecycle →I run unit and integration tests, verify artifact checksums, scan policy and dependencies, and block promotion when an evidence check fails.
Evidence: project CI/CD sections →I define calibration, drift, latency, and feed-health signals. SignalOps tests shadow evaluation, incident records, rollback, and reconciliation.
Evidence: shadow monitoring →I sign model evidence, chain audit events, check dataset rights, and verify artifacts and promotion policy before admission.
Evidence: registry and policy controls →I validate schemas and timestamps, reconcile sources, and bind point-in-time features and label availability to immutable manifests.
Evidence: data provenance path →I bring Python, SQL, REST services, algorithms, and full-stack engineering to this work, supported by degrees in computer science and information technology.
Evidence: resume ↓I wrote Terraform definitions for private Fargate tasks, TLS, scoped secrets, encrypted Multi-AZ PostgreSQL, and immutable evidence backups. CI checks the definitions. I have not deployed this AWS environment.
Review the deployment boundary →Master of Science in Information Technology
University of the Cumberlands · Graduated August 2026
View digital credential ↗Bachelor of Science in Computer Science
Idaho State University
View university verification ↗Full Stack Software Engineer
Tooroot Consulting Services · React, Node.js, Express.js, MongoDB, and REST interfaces.
MLOps Engineer | Machine Learning Engineer
I build reproducible ML pipelines, governed release gates, containerized APIs, model-supply-chain controls, and observable operations.