Matt Taylor ('27) carried out a detailed GIS suitability analysis for green infrastructure in the Krumkill Watershed. Using ArcGIS Pro, Matt built a multi-criteria evaluation model that incorporated factors such as drainage area, slope, imperviousness, hydrologic soil group, road buffers, and stream buffers to identify suitable locations for various GI types, including bioretention/rain gardens, vegetated filter strips, porous pavements, and infiltration trenches. Matt completed a full report for local partners, including the Town of Guilderland, to support their ongoing flood mitigation work at Stuyvesant Plaza and the Western Ave/Fuller Rd intersection. 

Veronica Choumarov ('27) examined climate vulnerability and historical injustices in the City of Albany and their relationship with a targeted case study in Albany’s South End, inspired by the community advocacy goals of removing I-787. She analyzed how historic redlining left the  South End vulnerable to flooding due to high impervious surface cover and proximity to I-787. Utilizing the Model My Watershed tool, Veronica simulated a design storm across four scenarios: replacing I-787 entirely with green space, or replacing the top 10 vacant properties in the South End with tiny forests, rain gardens, or wetlands. Her results show that replacing I-787 with green space produces the greatest flood mitigation benefits. 

Matt Taylor ('27) conducted a qualitative thematic analysis of 30 university AI policies using qualitative coding software (Delve and Taguette). He read and coded (identified patterns in the content) the AI policies and organized the codes into three main themes: Academics/Integrity, Administration, and Security/Legal. Matt also used VOSviewer to map the relationships between the 14 most prevalent codes (such as academic integrity, data privacy, transparency, and responsibility). A key finding of this study was the identification of a major policy gap: none of the 30 analyzed university policies referenced the environmental or energy impacts of AI usage.

Max Vanderbeek ('27) developed a “prompt router” that evaluates how hard a user's question is based on the prompt's length, reading level, and key reasoning words, and directs it to the smallest, most energy-efficient model capable of answering it. He tested the prompt router using nearly 20,000 real-world prompts. Four strategies were tested: always using the biggest model, picking a model at random, using a basic low-power router, and using an upgraded, fine-tuned router. Compared to always using the biggest model, random selection provided minimal benefit, the basic router reduced energy consumption by 37%, and the fine-tuned router cut it by 53%. In these latter two scenarios, answer quality remained nearly identical to that of the largest model. Results show that directing questions to the right AI model saves significant energy compared to relying on the largest model for every request.