Professional & research work
AgriLife Research
Tools and research workflows within Texas A&M Biological and Agricultural Engineering at AgriLife Research.
- My role
- Undergraduate Research Assistant, GIS
- Context
- Texas A&M AgriLife Research · Biological and Agricultural Engineering · Aug 2026–Present
- Toolkit
- ArcGIS Pro · Survey123 · Python · Tauri · PostgreSQL
- My contribution
- Geospatial analysis and research-tool development across Cotton Trax, PV-GIN, and cattle GIS.
- Evidence shown
- Research prototypes and dated documentation. Architecture figures distinguish intended integrations from demonstrated workflows.
Field evidence should survive a lost connection
Cotton Trax investigates traceability across harvest, transport, gin staging, and processing. Producer machines and yard scanners may be disconnected, and observations can arrive before the files that explain their origin.
My work combines architecture, software and GIS integration analysis, coordination, and documentation. The October 2026 handoff records a narrow Windows offline-producer MVP merged into main: bounded machine-file imports, original-byte preservation, validation, local SQLite records, duplicate handling, and a transactional outbox.
A separate local synchronization milestone demonstrated imports and durable receipts against Fastify and PostgreSQL. Shared Azure/ArcGIS integration, physical RFID validation, and full end-to-end traceability remained broader work. Lara’s frontend/dashboard work and Aiden’s hardware research are distinct team contributions.
Why an outbox and a separate GIS projection?
The Tauri/Rust producer app uses SQLite so original source evidence and pending synchronization operations can be committed together. Retry-safe server receipts distinguish a local save from central acknowledgement.
PostgreSQL/PostGIS is the intended canonical operational store. ArcGIS exposes rebuildable spatial projections rather than becoming the business source of truth. The first import’s 176 working records are source assertions, not 176 independently verified physical modules.
Make a weighted decision understandable
PV-GIN is a separate solar-feasibility initiative. Building on Wyatt’s original Survey123 survey, my contribution extended the V2 workbook with weighted comparison matrices, advanced weight controls, solar-finance displays, and publishing/presentation work.
Two independent six-criterion matrices compare process interference, parcel size, shading, soiling risk, terrain, and serviceability. Advanced mode allows weights to change in five-percentage-point increments, with feedback showing whether each matrix totals 100 percent.
The useful interface problem is making the weighting visible before a result appears. Survey testing and publishing are described in the October speaking notes; later dashboard readability requests are separate follow-up work. Related Beltwide 2027 author/proposal preparation is documented, without a claim of acceptance.
Survey123 implementation detail
The documented workflow uses XLSForm survey, choices, and settings tabs in Survey123 Connect. Formula placement, choice definitions, relevance conditions, and output spacing were part of the iteration.
ArcGIS Online and Enterprise DEV are separate publication targets. A successful publication in one portal does not establish the same form exists in the other. The archive preserves the written workflow rather than the final workbook formulas.
Turn a large raster into inspectable candidates
The cattle research workflow explores RAP raster data and candidate grazing polygons for the Boyd/McGregor work. Windowed raster retrieval limits processing to relevant areas, followed by local ArcGIS Pro classification and geometry cleanup.
A key debugging step was materializing Band 4 with Copy Raster before masking and thresholding. A control cell then retained its expected source value, making later candidate generation inspectable rather than relying on a temporary layer’s appearance.
An early rectangular mechanics test retained 13 candidates. Later October speaking notes describe 38 candidates from a September McGregor run and publication as an Enterprise DEV feature layer. Those are different runs; neither count is a field-validated grazing recommendation.
From raster values to usable geometry
The documented chain includes a single-band raster, AOI masking, threshold classification, raster-to-polygon conversion, projected area calculations, narrow-corridor cleanup, small-hole removal, and singlepart geometry.
The later workflow published vector features while hosted imagery publication remained blocked. Water, fences, terrain, and domain review are still needed before operational suitability can be established.
Different tools. Shared responsibility for evidence
These initiatives sit within my current research employment, but solve different problems. Cotton Trax concerns source evidence and synchronization; PV-GIN concerns decision support; the cattle work concerns geospatial processing and validation.
Across all three, my emphasis is on making assumptions, intermediate states, and technical handoffs visible enough for collaborators to inspect and continue the work.