# Statement of purpose — master draft

I want to develop trustworthy computational methods for observing and managing
living infrastructure: watersheds, stormwater networks, rail systems, and the
built environments that connect them. My path to this question is unusual but
coherent. I earned an M.S. and B.S. in economics at Georgia Tech, built a
sixteen-year career in computer vision and real-time perception, and repeatedly
encountered the same technical problem: consequential decisions must be made
from sparse, noisy observations whose meaning depends on relationships across
the whole system.

At Motion Reality, I developed real-time optical tracking and raised first-frame
marker identification accuracy from roughly 50 percent to 98 percent. At Beena
Vision Systems, I built a constraint-based identifier for railcar brake
components from wayside bogie imagery. Subsequent work included manufacturing
defect detection, surgical-navigation vision, and a licensed CT-registration
pipeline achieving approximately 0.5 mm validation accuracy. These systems
taught me to value physical constraints, auditable failure modes, and honest
uncertainty as much as headline predictive accuracy.

My independent research generalizes that lesson through hierarchical relaxation
labeling. Instead of classifying each observation in isolation, the method asks
whether the proposed labels are mutually consistent across pairwise and
higher-order relationships. A reproducible chirality experiment resolves 231 of
231 oriented cases with an order-three factor in a setting where pairwise
compatibility is provably blind. I now want to test and extend these ideas in
civil and environmental systems, not merely illustrate them in software.

My proposed research would fuse terrain, rainfall, water-level observations,
culvert or asset inventories, and camera/LiDAR data into uncertainty-aware
models of small watersheds and infrastructure. Conservation laws and network
topology can become explicit constraints rather than patterns a model must
rediscover from data. A parallel inspection direction would examine rare defects
in culverts, bridges, drainage systems, or rail assets, comparing constrained
inference with deep learning under missing data and distribution shift. My
economics training supports a third layer: evaluating how maintenance choices,
institutional incentives, affordability, and uneven risk shape adoption.

**[PROGRAM-SPECIFIC PARAGRAPH: Name two or three faculty; identify a laboratory,
dataset, or method; explain reciprocal fit without claiming prior contact.]**

I bring production engineering, quantitative social science, teaching, and a
record of carrying difficult systems from ambiguous requirements to validated
outputs. I also recognize the preparation demanded by civil engineering. Before
matriculation and during the first year, I am prepared to complete the required
background in fluid mechanics, hydraulics, mechanics, and domain-specific design
in consultation with the program. I seek a funded PhD environment where that
preparation supports rigorous, field-connected research.

My goal is to produce methods that researchers and practitioners can inspect,
challenge, and use: reproducible benchmarks, clear separations between measured
and inferred quantities, and tools that improve maintenance and environmental
decisions without pretending uncertainty has disappeared. **[PROGRAM]** offers
the combination of **[FACULTY/METHOD]**, **[FIELD SYSTEM]**, and **[PUBLIC
MISSION]** needed to turn that goal into a disciplined research program.