AI engineerArizona / Earth
I build what matters, understand why it works, and never stop making it better.
Selected transmissions · 2024-26
Production work, active bets, and experiments that kept moving after the demo.
Built Paperlyft.com to turn scanned PDFs into accessible documents while keeping the content, layout, and handwritten signatures. It uses multimodal AI and a custom agent harness to generate pages, check them, and repair issues. Automated accessibility checks, visual review, and human feedback are part of the workflow, and AWS handles the processing asynchronously.
Research-backed context and review tooling for Claude Code, with branch-local continuity, privacy-aware evidence, reusable review contracts, and an advisory GitHub Action path.
A nonprofit platform with CMS-driven pages, admin operations, dynamic forms, donations, and payments, built by a team to help families, professionals, and donors find and request services.
A project workspace where Kanban, documents, journals, and Studio intake are connected by a multi-agent LangGraph orchestration layer with durable state and role-aware workflows.
I use Final Destination to find and process jobs in batches from a browser extension, tailor resumes, and I just need to review and apply.
Method / operating principles
I build AI systems, check where they fail, and keep improving them so they work reliably in everyday use.
I use the same approach in Paperlyft.com and my code-review work at ASU. The problems are different, but in both I want evidence that the system does what it claims.
A well-scoped agent with the right context usually beats a smarter model with no boundaries.
If a decision cannot be traced to an agent, context, and rule, the system is not production-ready.
One reliable capability is more valuable than ten impressive behaviors that drift between runs.
Trajectory
I’m building the document conversion system and its AWS infrastructure with CDK, Lambda, ECS Fargate, SQS, S3, and DynamoDB. I use model selection, token-level cost tracking, and queue controls to balance output quality, latency, and cost. I also work with county election offices and prospective customers, and use their feedback to decide what to build and how the system should work.
Designing the orchestration, memory, authority, and infrastructure layers for reliable agent teams.
Working instruments
AWS Certified Cloud Practitioner · AWS Academy Graduate - Cloud Security Foundations Training Badge
Interests / future work
My work is rooted in education, while my curiosity keeps pulling me toward finance and the edges of computation.
Currently working at the Learning Engineering Institute to help change the world of education - building technology that makes learning more accessible, more useful, and more human.
I’m drawn to finance for the way it combines data, uncertainty, incentives, and long-horizon decisions. I want to keep learning the systems behind the numbers and eventually build within them.
Quantum computing interests me because it asks us to rethink what computation can be. I’m learning about the field simply because the ideas are fascinating - and to see where they may lead.
Different domains. The same question: how can thoughtful systems expand what people are able to do?
Field notes
