Work

Forward-Deployed Engineering. Embedded with doctors, finance, and ops.

As the only engineer at a dental implant company, I sit with the person who has the problem, ship into their tools, then train them.

FDE 7 projects · 5 articles

How I work.

No one writes tickets. My users are surgeons, a CFO, treatment coordinators and a chief sales officer. None of them write tickets.
Discovery is the work. Requirements live in their spreadsheets, meeting transcripts and workarounds, so that is where I dig.
Build where they work. I rebuilt leadership's Excel dashboard in Google Sheets, iterated live with the CSO, every number independently checked.
Fix the data first. Fixing the broken patient to lead links surfaced about $460K of patient value the old reports dropped.
Teach, then step back. A weekly AI Power Hour and office hours for seven nontechnical departments, built from a needs survey.
The honest no. My CRM blueprint tells leadership plainly: without funded engineering, stay on the current vendor.

Discovery from primary sources

Requirements from real exports, transcripts and shadowing: eight data problems catalogued with row counts.
Ship into existing workflows

Sheets for ops, a web UI for finance, email reports for doctors.
Enablement and training

Weekly training and office hours that get nontechnical teams to confident, compliant AI use.
Honest scoping

Build versus buy analyses with real costs, including the ones that say don't build.

In production

One row per patient, every document one click away, in a practice where every system named patients differently.

Patients, intake surveys, manual intake, patient experience, estimates and consultations in one place
Records now join on the Denticon id, and a person's link decision always beats the matcher
Data freshness went from six days behind to 2.6 minutes behind once the sync ran on a timer (measured on staging)

In production

Replaced a seven sheet staffing workbook with one forward view of demand and capacity across three offices.

Started from the workbook itself, including the fixed ranges that undercounted without a warning
Burnout treated as a design goal: proposed defaults flag three heavy days in a row
Cover suggestions explain themselves: five ranked keys, every reason traced to its rule

In build

A diagnostic system built with the practice founder that unifies every imaging type into one doctor validated report.

In production

The operations dashboard leadership runs the business on: weekly Monday.com exports turned into a full funnel view.

One definition per metric since the September rebuild, so a card and its drilldown cannot disagree
Surfaced ~$460K of orphan patient value the old reports dropped
Ambiguous patient to lead links cut from 31% to 5.8% with first name checks, recorded per row

Running program

Weekly AI Power Hour sessions and open office hours for seven nontechnical departments, built from a needs survey.

Design complete

A ready to build architecture for moving 15K patient records off Monday.com, with a marketing firewall that can't leak PHI.

PRD says the quiet part out loud: without funded maintenance, don't build this
Human-in-the-loop dedup; reversible merges; audit log captures reads, not just writes
NY SHIELD analysis citing a $2.85M settlement in Hybridge's own market