what i do · applied ai, data, and software

Built plainly, documented, handed over

Most of what I take on starts as a workflow that has outgrown its spreadsheet. The technical half is the easier half. The other half is understanding how the decision actually gets made and where a tool can quietly remove friction.

AI and data tools

Recommendation systems, decision-support workflows, and analytics that answer a question someone actually asks. Where a model is involved, a person stays accountable for the decision and the reasoning is legible.

LLM workflowsrecommendation systemsembeddingsmodel evaluationhuman in the loop

Dashboards and reporting

Reporting views an owner can read without a translator. Built on the numbers that drive a decision, not every number the system happens to store.

PythonSQLpandasStreamlitdashboard design

Custom applications

Booking flows, intake, client communication, internal tools. Built plainly, on a stack that will still make sense to whoever picks it up next.

FastAPIReact/Next.jsPostgreSQLSupabaseAPIs

Scoping and strategy

Sometimes the answer is that you do not need software yet. Watching how the work actually moves through a business comes before automating any of it.

requirementsworkflow modelingprocess improvementAI strategygovernance
  • Work where you would not own the result. The source, the data, and the docs go with you.
  • Automating a process nobody has fixed yet. That makes a broken process faster, not better.
  • A model where a rule would do. If deterministic code can make the call, it should.
  • Numbers I cannot source. No invented metrics, on this site or in a deliverable.

source: the standards carried over from enterprise internal-systems work

Next

If that sounds like the shape of your problem, the next step is a conversation about the workflow, not a proposal about the software.

Start a project