Dylan Albert
Full Stack & AI Engineer
I build full stack applications, cloud infrastructure, and the AI workflows that run on top of them.
About
I'm strongest at taking existing systems and making them measurably better, from retrieval through evaluation to observability.
Full Stack Cloud Engineer with nine years of experience, fully remote since 2020. I've shipped production AI systems end-to-end, owning everything from retrieval and evaluation to multi-agent orchestration and observability, and have extensive multi-cloud and DevOps experience. At Oracle I drove a ~30% average correctness lift on production RAG applications. At IXOPAY I rapidly prototyped and built the core agents of an agentic SDLC pipeline that turns JIRA work items into design documents, code changes, and review-ready PRs.
Hands-on with Python, Java, Kubernetes, LangChain and LangGraph, and three major cloud providers (OCI, AWS, and Microsoft Azure), spanning the full AI pipeline from retrieval through evaluation to observability. Known for unblocking teammates, mentoring across seniority levels, and moving fast from zero to prototype to production. I've built most of it end-to-end: KM apps, RAG evaluation frameworks, multi-tenant internal AI platforms, CI/CD pipelines, and backend APIs powering shared services.
- Focus
- RAG · LLM evaluation · Multi-agent orchestration
- Currently
- Product Engineer at an undisclosed government contractor
- Stack
- Python · Java · Kubernetes · OCI / Azure / AWS
- Based in
- Tulsa, OK · remote since 2020
Mentorship
At Oracle I became the de facto onboarding and mentorship anchor as the AI Services team scaled, supporting eight engineers from intern to senior on day-to-day feature work, troubleshooting, and planning.
Career Journey
Started in occupational health & safety before pivoting hard into software. That background (systems thinking, risk analysis, process improvement) turned out to be a genuine asset in AI engineering, where you're constantly reasoning about failure modes, data quality, and safety.
Experience
Product Engineer
Details of this engagement are not publicly disclosed.
AI Engineer
Rapidly prototyped and built the core agents of an agentic SDLC pipeline that turns JIRA work items into design documents, code changes, and review-ready PRs, and provisioned the Azure environment that hosts it.
- Rapidly prototyped and built the core agents of an agentic SDLC pipeline that turns JIRA work items into design documents, implements code changes, and drafts review-ready PRs, with multi-model consensus checks and operator checkpoints along the way. Owned the planning, architecture, and documentation agents end to end, built the frontend agent and backend agent, and contributed heavily to the shared agent library.
- Provisioned a new Azure environment from scratch using Terraform / OpenTofu (CloudNation modules) and Azure DevOps to host the agent pipeline's router service: VNet, Container Apps Environment, Container App, Key Vault, dual managed identities (a federated UAMI for pipeline auth and a service UAMI for runtime Key Vault access), and Application Insights integration. Stood up end-to-end in roughly four days, moving the router off a developer workstation to a monitored environment with single-command revision rollouts.
Cloud Software Developer
Drove measurable quality lifts on Oracle's production RAG applications and served as the de facto onboarding anchor as the AI Services team scaled.
- Improved answer correctness on existing GBU RAG applications by ~30% on average (roughly 60% to 90% across the evaluation dataset, measured against ground-truth data via LLM-as-a-Judge, with per-application variance tied to source-documentation quality). Drove the improvements through semantic chunking optimized for retrieval precision, summary generation with cross-chunk overlap for richer generation context, chunk-level metadata enrichment (e.g., product version tagging) to improve retrieval filtering, and post-retrieval re-ranking.
- Served as the de facto onboarding and mentorship anchor as the AI Services team scaled, supporting eight engineers ranging from intern to senior on day-to-day feature work, troubleshooting, and planning while delivering on my own roadmap.
- Designed and delivered a Knowledge Management AI application that transforms customer-support interactions into structured KB articles (title, cause, symptoms, solution) for a customer-facing portal, built end-to-end using a structured-output pipeline with OCI guardrails and PII redaction applied before LLM invocation.
- Deployed and heavily customized Open Web UI as a multi-tenant internal platform for GBU RAG applications, building a suite of custom pipe/routing logic (SSE streaming, query rewriting, product-based routing, tailored system prompts per product) that functioned as an inline multi-step agent workflow across multiple business units.
- Built a RAG evaluation framework from scratch that ran queries against golden datasets and produced per-dimension metrics (relevancy, correctness, completeness, context faithfulness, hallucination rate), then drove the team's migration to Langfuse once it matured to provide both observability and LLM-as-a-Judge in a single platform.
Cloud Software Developer
Primary engineer on shared backend services consumed across Oracle's Global Business Units, including two Notifications services built from scratch under Principal-Engineer design review.
- Led end-to-end implementation of two Notifications services built from scratch (one general-purpose, one for a specific GBU use case), owning Java/Micronaut backend APIs, data models, and OCI integrations under design review from a Principal Engineer.
- Delivered features and improvements to additional shared services including Smart-Search and Redwood Telemetry, part of a platform consumed across Oracle's Global Business Units.
- Integrated services with OCI primitives (Oracle DB, Object Storage, Events, Streams, Distributed Tracing, APM) to deliver scalable, cloud-native backends.
- Built and maintained GitLab CI/CD pipelines automating builds, testing, scanning, deployments, and release notes, reducing manual intervention between releases.
- Deployed and maintained containerized applications via Kubernetes and Docker, expanding unit, integration, and performance test coverage to raise release confidence across services.
- Mentored junior developers and interns on debugging, clean-code practices, and Micronaut application development.
Full-Stack Application Developer
Replaced paper-based administrative processes with full-stack applications serving thousands of students, high schools, and internal staff across the Tulsa metro area.
- Designed and shipped the Concurrent Enrollment application, fully replacing a paper-based enrollment process used across every public high school in the Tulsa metro area. The app coordinated a multi-stakeholder workflow (student request, parent approval, high school counselor review, TCC staff enrollment) for thousands of juniors and seniors taking TCC courses each year.
- Built an HR performance evaluation app and a Grant Management system, each replacing long-running manual processes and serving internal staff across multiple departments.
- Delivered full-stack applications using Python, PHP, and Java with complex SQL integrations across Oracle, MySQL, and Postgres databases, and Oracle APEX / Vanilla JS / Tailwind / PL/SQL on the frontend and backend.
- Refactored and modernized legacy codebases, improving maintainability and performance across public-facing and internal applications.
Skills
Tap any skill for where and how I've used it.
AI / ML
Languages & Frameworks
Databases
Cloud
DevOps & Infrastructure
Project Highlights
Key initiatives and measurable outcomes.
Agentic SDLC Pipeline
Built at IXOPAY, Mar 2026 to Aug 2026Challenge
Turning a single JIRA work item into a shipped, review-ready change requires context from across the repo, consistency with existing architecture, and validation at every step before a human signs off.
Outcome
Replaces ad-hoc planning and hand-off with structured, validated design docs and review-ready PRs, shrinking the gap between "work item filed" and "ready to merge."
Approach
Rapidly prototyped and built the core agents of the pipeline. The Planner agent ingests JIRA work items, cross-references repo documentation and in-scope code, and produces a structured design document with file-level impact analysis, sequenced user stories, merge ordering, and parallelization recommendations. Frontend and backend agents implement the code changes and draft review-ready PRs. Complementary Architecture and README & Dependencies agents generate ARCHITECTURE.md files, ADRs, Mermaid data-flow diagrams, and dependency docs as upstream context. Multi-model consensus checks and operator checkpoints gate each stage before a human ever sees the output. Owned the planning, architecture, and documentation agents end to end, built the frontend and backend agents, and contributed heavily to the shared agent library.
Pipeline
- 1Context
JIRA work item + repo
- 2Plan
structured design doc
- 3Implement
frontend + backend agents
- 4Validate
multi-model consensus
- 5Review
review-ready PR, operator checkpoint
Context Agents Feeding the Planner
- Architecture agent: generates ARCHITECTURE.md files, ADRs, and Mermaid data-flow diagrams
- README & Dependencies agent: generates dependency documentation and READMEs
RAG Correctness Overhaul
Challenge
Oracle's production GBU RAG applications had limited, inconsistently-measured correctness across a wide product surface, and generation quality varied heavily by source-documentation quality.
Outcome
~30% average correctness lift (roughly 60% → 90% across the evaluation dataset), with per-application variance traceable back to source-documentation quality, giving the team a measurable, defensible baseline for further investment.
Approach
Drove improvements through semantic chunking optimized for retrieval precision, summary generation with cross-chunk overlap for richer generation context, chunk-level metadata enrichment (e.g., product version tagging) to improve retrieval filtering, and post-retrieval re-ranking. Measured against ground-truth data via LLM-as-a-Judge with per-application variance surfaced in the report.
Knowledge Management AI App
Challenge
Support teams lacked a scalable way to convert customer-support interactions into reusable, structured knowledge base articles.
Outcome
Production application powering a customer-facing portal, directly accelerating support workflows and converting one-off interactions into reusable knowledge.
Approach
Designed and delivered the application end-to-end using a structured-output pipeline that emits title, cause, symptoms, and solution fields per KB article. OCI guardrails and PII redaction are applied before LLM invocation to keep customer data out of the model call.
RAG Evaluation Framework → Langfuse
Challenge
No systematic, repeatable way to measure RAG quality across Oracle's product surface, and improvements were hard to justify without a data-driven baseline.
Outcome
Repeatable, data-driven quality baseline enabling ongoing RAG improvement with measurable benchmarks, now unified with tracing and observability in a single platform.
Approach
Owned end-to-end. Built a golden-dataset testing framework producing per-dimension metrics (relevancy, correctness, completeness, context faithfulness, hallucination rate) across multiple RAG APIs and OCI-hosted LLMs. Drove the team's migration to Langfuse once it matured, unifying observability and LLM-as-a-Judge in one platform.
Multi-Tenant Open Web UI Platform
Challenge
Multiple Oracle business units needed a unified, configurable AI interface integrated with internal RAG services, and each BU had its own routing and prompt needs.
Outcome
Internal AI platform serving multiple GBU products, enabling self-serve AI experimentation and production use across business units from a single deployment.
Approach
Forked and heavily customized Open Web UI as a multi-tenant internal platform. Built a suite of custom pipe/routing logic (SSE streaming, query rewriting, product-based routing, tailored system prompts per product) that functioned as an inline multi-step agent workflow on top of the base UI.
Concurrent Enrollment Platform
Still in production at Tulsa Community CollegeChallenge
Tulsa Community College's dual-credit program enrolls high school juniors and seniors in TCC courses for college credit, and every step of that enrollment ran on paper. Enrollments routed through a multi-stakeholder chain: student picks classes that fit their schedule (not officially enrolled yet), parent approves, high school counselor reviews, TCC dual-credit staff finalizes enrollment. Per-student cycle time was weeks to months, and the manual paperwork load on dual-credit staff was brutal.
Outcome
Rolled out across every public high school in the Tulsa metro area, serving thousands of juniors and seniors each year. Per-student cycle time collapsed from weeks or months down to days. Dual-credit staff moved from retyping paperwork to actually processing enrollments. Still in production six years later.
Approach
Designed and shipped an end-to-end web app coordinating the full workflow digitally. Oracle APEX with PL/SQL as the backend (yes, PL/SQL as application logic, which is an adventure), plus Vanilla JS and Tailwind on the front end. Every step (student course selection, parent consent, HS counselor review, TCC staff enrollment) happens in the app, with status visibility at every hand-off.
Workflow
- 1Select
student picks classes that fit their schedule
- 2Approve
parent consent captured
- 3Review
HS counselor signs off
- 4Enroll
TCC dual-credit staff finalizes
Education & Honors
Auburn University
B.S. Computer Science
University of Central Oklahoma
B.S. Occupational Health & Safety
Honors & Awards
Let's build something
Open to Full Stack, Cloud, and AI Engineering roles. The fastest way to reach me is email; I read everything.