Mark Kenneth Badilla

Hi, I'm Mark

Full Stack AI Engineer: Hand me a goal instead of a spec and I come back with the diagnosis, the recommendation and the system.

MKB

Live agent, built for this site

Ask it something, and if the answer is on the page it will take you to it.

About

What I own: AI and data automation at WeAssist, end to end.
What arrived: A goal rather than a backlog.
What I built first: The picture of how work really moved between departments and the tools they already had, written up with a buy-versus-build recommendation.

I argued for buying or consolidating wherever something adequate existed, and reserved custom work for the one gap nothing covered.
I built the internal operations platform (Next.js and PostgreSQL) in ten weeks, on additive-only migrations.
A six-area permission grid that has no bypass role by design.
Its own gate suite is wired into the build, so bad code is uncommittable whether a person or an agent wrote it.
I self-host with Coolify, which lets us run anything at one flat rate.
Every deploy backs up the database, restores it into a scratch database and diffs the row counts before anything migrates.
Also built a meeting platform used daily.
An unattended LLM extraction pipeline with a ranked plain-language risk worklist on top.
Desktop AI agents for non-technical staff.
A label audit showed the data could not support an honest predictive model.
I killed that path on its own evidence and shipped the explainable rules-based detector instead.

Before that: Core systems for a multi-tenant ERP platform at Hatchit Solutions, where one bug is every client's bug at once.
Degree: BS Information Technology, Magna Cum Laude

Own AI and data automation end to end.
Built the internal operations platform (Next.js and PostgreSQL) in ten weeks.
Access control rebuilt from 55 scattered permissions into a six-area grid with no bypass role by design.
Its own gate suite of 300+ checks wired into the build.
Bad code is uncommittable whether a person or an agent wrote it.
Self-hosted with Coolify, so we can run anything at one flat rate.
One 18-step zero-downtime deploy pipeline every app shares.
Every deploy backs up production, restores it into a scratch database and diffs the row counts before anything migrates.
Built an unattended LLM extraction pipeline that turns unstructured meeting transcripts into a structured signals database.
Then shipped a rules-based, explainable scorer that ranks records by risk with the reason written out.
Killed my own flagship predictive-model project when a label audit showed the data could not support an honest model.
Shipped an explainable rules-based detector instead.