Principal AI/ML Architect — retrieval evaluation, governed agent runtimes, and enterprise AI architecture.

I build AI systems that are inspectable, citation-grounded, and audit-friendly. The public repos below are reproducible reference implementations; each README states what it does and does not cover.

• production-rag-eval-harness — Reproducible local harness comparing dense, sparse, hybrid, and graph-aware retrieval over a public corpus, with citation-grounded answers, a committed scored run, and reproducibility manifests. Regression gate on the roadmap.
• agent-runtime-observability — Governed agent runtime reference: bounded retries, policy gates that deny unsafe tool calls, a documented failure-mode catalog with reproducible triggers, OpenTelemetry-shaped JSON traces, a five-tool layer with input-schema contracts, and a synthetic fixture corpus exercised through a deterministic stub LLM. Recorded runs are committed on main: canonical demo + seven policy-gate scenarios + five failure-mode triggers.
• aws-bedrock-iac-reference — Bedrock-anchored AWS reference architecture as Terraform IaC, surfacing security, cost, observability, and cloud-hygiene judgment. Dry-run evidence is committed on main: terraform validate + plan + tfsec + checkov outputs under plans/canonical/.

• Retrieval evaluation discipline: scored local runs, citation contracts, reproducibility manifests.
• Governed agent runtime design: policy gates, failure modes, traceability, bounded tool use.
• Enterprise AI architecture judgment: AWS / Bedrock, security, cost, and operations posture.
• Honest evidence boundaries: these repos do not claim large-scale inference ownership, RLHF / DPO / LoRA training, MCP server delivery, production SaaS deployment, or customer deployment proof.

• Marine Corps Senior Intelligence Analyst (2015–2020).
• BS Cyber Engineering, Houston Christian University (2024).

LinkedIn: in/ariel-lee-4a6a231aa Email: ariel.j.lee@outlook.com Location: Houston, Texas