Selected Work

Work that turns ambiguous physical and technical problems into testable systems.

I am most useful when a team has a difficult object, too many plausible directions, and no clean way to tell what is working. I help find the core decision, make it inspectable, and build the next test.

Advanced Metal Research

Constraint: a useful weld planner has to look ahead through awkward geometry without producing jerky or unsafe robot motion.

Work: a test-driven planning and search harness for weld sequences and robot paths, with explicit representations for welds, waypoints, poses, and candidate configurations.

Evaluation: smoothness, jerk, reachability, collision risk, torch angle, configuration changes, cycle time, and similarity to manually programmed paths.

Current state: active project. The first milestone is a bounded offline harness that generates and compares several candidate plans on real CAD and CAM examples.

Applied Medical

Constraint: tooling had to survive material variation, operator habits, machine timing, and production pressure.

Work: CAD, prototypes, fixtures, pneumatic tooling, robotic EOAT, test, qualification, and shop-floor debugging.

Outcome: a MIM insert-molding process fix cut roughly $100k/month in operating variance.

Mechanical details

T-UEBA

Constraint: tactical behavior shifts fast. Labels stay sparse. Analysts need evidence, not alert volume.

Work: constrained heterogeneous temporal graphs, calibration, active learning, synthetic-data triggers, and evidence drilldown.

Outcome: $250k SBIR PI work for Army tactical Zero Trust behavior analytics.

Cognitive security · RAM Labs

DXF/CAD representation

Constraint: geometry, relations, editability, and auditability have to survive conversion.

Work: exact artifact, object state, typed relation graph, deterministic validators, learned residual repair.

Signal: HATCH-IoU, provenance, and reviewable state make CAD-native learning less vague.

Graph modeling · Applied epistemics

AIxCC / DL-Patcher

Constraint: code repair needs checker pressure, runtime feedback, and human review. Prompting alone cannot carry the patch loop.

Work: code-model harnesses, static analysis, fuzzing, verifier feedback, and patent-filed repair architecture.

Agent environments

Production systems

Constraint: small teams need systems that stay legible under cost, latency, and operational limits.

Work: AWS workflows, thermal/GIS processing, customer-facing tools, and production handoff.

Resume variants