Jorge Antonio A.

I design, build and improve business systems: custom applications, data pipelines and workflows that connect financial planning with day-to-day operations. That includes AI automation and agents where they fit. I work from the problem outward: understand how the work happens today, where information breaks down and what improvement would matter. Then we choose the simplest suitable approach and agree how success will be checked. What I do: • AI automation and agents, new or existing: build, repair or extend workflows around your business tools. Before building an agent, I check whether plain code, a conventional model, one model call or a fixed workflow would meet the goal. We start with a one-page brief (the user, task, trigger, connected systems, cost and timing limits, permitted actions, approval points, success criteria and unacceptable failures). Where an agent looks warranted, I plan a fair comparison with the strongest practical fixed workflow before committing to model-directed control. • Integrations with least access: connectors to your files, databases and business tools through MCP, the Claude and OpenAI APIs and Azure, giving the AI only the access it needs. • Testing, new or existing systems: representative cases with agreed pass criteria, known answers where available, and reviewed synthetic cases to explore gaps. We record what the tests establish and what still needs validation with real data. • Audit and rollout: action logs, staged release with rollback, monitoring, and a named owner who turns failures into regression tests. • Finance and data: xP&A, financial modeling, business valuation, data pipelines and Power BI on Microsoft Azure. In AI-assisted financial workflows, calculations and forecasts come from testable code or models; the AI helps gather inputs and explain results. After a decade in finance, I got tired of seeing important decisions made with disconnected data, so I started building the bridge myself. Since 2022 I have pursued independent research into AI and large language models, focusing on their practical capabilities, limitations and appropriate use in business. I wrote an open design process for agentic systems (nine phases, each with a "done when" gate) and a worksheet for deciding whether a task needs an agent at all; both are public on GitHub. I also analysed about 220 published requirements for AI systems that use software tools. In 39%, no success check was stated, a gap I address by agreeing acceptance criteria before building. Credentials: MS, Big Data & Analytics (BarcelonaTech); MS, Accountancy – Data Analytics (University of Illinois Urbana-Champaign); Advanced Valuation Certification (NYU Stern Executive Education). Bilingual English/Spanish. Tell me what the system should do and how you would judge it; I will recommend the simplest approach that does the job and the tests needed to confirm it.