Artificial Intelligence (AI) meeting assistants have increasingly been adopted to support online meetings by automating tasks such as note-taking, meeting summarization, and action item extraction. By leveraging advances in natural language processing and speech recognition, these tools help reduce cognitive load and improve meeting productivity. However, the wide range of features provided by a single AI meeting assistant makes it difficult to recognize the value of each. This study aims to explore the key functionalities, benefits, and challenges users face when interacting with these tools. Through survey responses, our goal is to understand how to design AI meeting assistants to better support the user's workflow. We conclude with a categorized breakdown of the positive and negative aspects identified by participants as well as a list of requests and suggestions from participants. Understanding how users interact with AI meeting assistants is essential for designing a tool that will deliver its promise of supporting productivity in online meetings. This study demonstrates how user-centered evaluation of AI meeting assistants can inform both design and deployment strategies, ultimately contributing to a more effective and efficient collaborative work environment.

As atmospheric levels of CO2 continue to rise, direct air capture (DAC) technologies become increasingly crucial to mitigate the impacts of climate change. However, the high energy demand associated with sorbent regeneration remains a barrier to large-scale DAC implementation. Microwave (MW) heating offers a promising route to reduce the energy required for sorbent regeneration by offering rapid, selective volumetric heating compared to slow conduction and convection-based heating methods. Therefore, this project investigates the impact of incorporating microwave-active additives into hybrid CO2 capture materials on the sorbent’s overall MW regeneration performance. To test this, Nanoparticle Organic Hybrid Materials (NOHM) embedded in a solvent-impregnated polymer matrix (SIP) were synthesized with MW-active materials such as carbon black nanoparticles and Fe₃O₄ powder. Fourier-transform infrared spectroscopy was used to confirm structural integrity, while thermogravimetric analysis evaluated CO₂ adsorption capacity. It is expected that sorbents containing MW-active additives with high dielectric loss will regenerate more rapidly with lower electrical energy input during MW regeneration compared to conventional regeneration, while maintaining comparable adsorption capacity. This work is significant because it will guide the design of more energy-efficient sorbents for MW regeneration and scalable carbon capture.

Autonomous robotic pollination requires high-fidelity 3D perception to navigate complex, unstructured environments without damaging delicate plant structures. This research addresses the challenge of creating accurate plant reconstructions for collision-free path planning using an AgileX Piper robotic arm. While initial attempts utilized fixed-position cameras, this study pivoted to an eye-in-hand configuration to overcome occlusion, necessitating custom hardware integration and hand-eye calibration using the Daniilidis AX=XB solver. The core methodology involved an autonomous "plant-aware" reconstruction pipeline: utilizing Grounding DINO and Segment Anything Model (SAM) to ensure target plants were optimally centered and fully contained within the frame before point cloud capture. These filtered clouds were fused via Iterative Closest Point (ICP) to improve alignment and establish a clean static model. For execution, an open-loop path planning flow was implemented in MoveIt, where Octomap tuning and the suspension of map updates during arm movement were critical to prevent "smearing" and ensure a reliable collision environment. This work establishes a robust foundation for future real-time adaptive strategies, including Grasp-net pose selection, SE(3) retargeting to handle plant junction drift, and the integration of adaptive IBVS or SARSA at the final approach phase to ensure precision in dynamic agricultural environments.

Conventional cement production accounts for 8% of global CO2 emissions, primarily due to limestone decomposition. This project investigates a green cement pathway by electrochemically producing Ca(OH)2, a key cement precursor, using seawater as an abundant resource. The goal is to generate acid and base through electrolysis, and use them to dissolve limestone and form Ca(OH)2, while reducing CO2 emissions. Our flow electrolyzer consists of two compartments separated by an ion-selective membrane, with an anode and cathode driven by an applied voltage. Electrolysis of a saline feed produces H+ at the anode and OH- at the cathode: acid dissolves CaCO3 to form CaCl2, while base generates NaOH, used to produce Ca(OH)2. A competing Cl2 evolution reaction at the anode reduces efficiency and must be minimized. We evaluated Ca(OH)2 production by measuring current efficiency and Cl2 evolution to determine how effectively the applied current produces OH- and to identify losses from side reactions. Current efficiency was calculated using applied current, flow rate, and OH- concentration measured through titration. To optimize the system, we tested ion-selective membranes, including anion exchange membranes (AEM) and cation exchange membranes (CEM). AEM showed higher efficiency and better control of side reactions. These results highlight membrane selection as critical for improving efficiency and achieving a scalable low CO2 pathway for green cement production. Future work will focus on further optimizing cell design & membranes.

Lateral-flow immunoassays (LFAs) are commonly used to detect proteins in bodily fluids at the point of care (POC). Although commercially successful, LFAs suffer from low sensitivity. To improve their sensitivity, our group was the first to implement polymer-salt aqueous two-phase systems (ATPSs) to preconcentrate various targets into a smaller volume prior to LFA application. For small hydrophilic protein targets that distribute fairly evenly in polymer-salt ATPSs, polymer-polymer ATPSs may be used for preconcentration instead. A poly(ethylene glycol)-dextran ATPS can preconcentrate proteins into the dextran-rich bottom phase, but has not been integrated with LFAs due to poor sample flow caused by the highly viscous dextran-rich phase. To address this limitation, we implemented the enzyme dextranase (DN) to degrade dextran in the bottom phase sample, reducing its viscosity before LFA application. Enzymatic studies were performed and then further evaluated with imbibition studies to demonstrate improved fluid flow on paper. The ATPS was then combined with DN to achieve a 10-fold improvement in the limit of detection for the protein transferrin spiked in artificial saliva when compared to a conventional LFA setup. To improve POC accessibility, DN was dehydrated on paper and applied to the LFA to achieve the same improvement in sensitivity. This study represents the first successful use of a polymer-polymer ATPS to improve LFA sensitivity, and introduces the possibility of integrating a different class of ATPSs with the LFA.