Impact-Driven & Applied Research:
Collaborative Project Repository

This repository compiles DSSG applied research initiatives across Florida. It serves as an open record of how we combine AI and data science with nonprofit and public sector partnerships to solve real-world challenges.

DSSG Projects Repository Listings

Showing 20 Project(s)
Project ID: #AI4DSSG-020
September 2026 – Current Funded Project Ongoing Research Project Stipended Position

Mapping LIFT JAX Data Readiness for Impact Analysis

Technologies: Data Readiness AssessmentData Asset CatalogingExternal Landscape ScanningResearch Scoping & DesignCensus Data

Community Partner: LIFT JAX

Social Issue: Eradicating Generational Poverty

LIFT JAX coordinates comprehensive housing, education, wellness, and economic-vitality initiatives to eradicate generational poverty in Jacksonville’s Historic Eastside neighborhood. As the organization scales alongside major new public and private investments, leadership requires a rigorous independent assessment of its current data infrastructure before it can design a formal impact evaluation. Establishing LIFT JAX’s specific contribution to neighborhood change is a complex research challenge that cannot rely on internal program data alone. To plan long-term research and prove its direct attribution to neighborhood evolution, the organization must first identify internal tracking gaps and catalog external data sources that can serve as geographic counterfactual controls.

Research and Solution Development

This project is a newly initiated, semester-long scoping and readiness engagement. The research workflow is structured strictly to audit internal data assets across LIFT JAX's four core pillars, build an external neighborhood-level data landscape catalog, and establish forward-looking research questions. Detailed technical methodologies, data discovery catalogs, and research design frameworks will be developed and integrated as our weekly collaborative meetings progress.

Impact and Outcomes

The final success criteria for this short-term engagement will be the delivery of a comprehensive data asset inventory, an external neighborhood data catalog mapped to the Eastside footprint, and a written long-term research design report. These foundational scoping outputs will be handed off to LIFT JAX leadership and future FL-DSSG cohorts to eventually execute a credible, evidence-based attribution study.

Student Researchers
Brant Marcus (Psychology)
DSSG Experts: Dr. Dan Richard (Psychology), Dr. Karthikeyan Umapathy (Computing)

 

Project ID: #AI4DSSG-019
June 2026 – Current Pro-Bono / Unfunded Ongoing Research Project Independent Study

The Economic Impact of Improving Life Expectancy in Jacksonville

Technologies: Economic Valuation FrameworksPublic Health Data SynthesisPythonmySidewalkTableauAdvanced Statistical Analysis

Community Partner: State of JAX Initiative, City of Jacksonville

Social Issue: Life Expectancy

Life expectancy in Duval County trails behind both Florida and the national average, varying by as much as 25 years across different Jacksonville neighborhoods. The city lacks a simple framework to link these local health disparities to social determinants like housing, employment, and healthcare access. To help policymakers justify community investments, a method is needed to map these neighborhood differences and calculate the economic value of improving local life expectancy.

Research and Solution Development

This project is currently a research initiative in its beginning stages. The upcoming work focuses on creating a local Social Deprivation Index (SDI) at the county and census-tract level; and outlining a framework to calculate economic value using standard statistical life-year methods. Detailed data processing models and pipeline tools will be shared later when the research design is finalized.

Impact and Outcomes

This project is currently in progress, and no data analysis or modeling has been conducted yet. The final goal is to estimate the economic return of a one-year improvement in Jacksonville's median life expectancy. Specific findings, metrics, and policy recommendations will be added to this section at a later date once the research and analysis are complete.

Student Researchers
Arjun Vooturi (Data Science)
DSSG Experts: Dr. Parvez Ahmed (City of Jacksonville), Dr. Nishi D’Souza (Public Health), Dr. C. Dominik Güss (Psychology), Dr. Dan Richard (Psychology), Dr. Tes Tuason (Public Health), Dr. Karthikeyan Umapathy (Computing), Sneha Kattari (Data Science)

 

Project ID: #AI4DSSG-018
August 2026 – Current Pro-Bono / Unfunded Ongoing AI Innovations Research Experience

Privacy-Aware Multimodal Retrieval for Wish-Fulfillment Photo Archives: AI Enrichment and Hybrid Semantic Search

Technologies: Vision-Language ModelingMultimodal Image EmbeddingsHybrid Semantic SearchHugging FacePython

Community Partner: Dreams Come True

Social Issue: Dream Fulfillment

Dreams Come True (DCT) has accumulated a large, multi-year photographic archive documenting more than 5,000 wish-fulfillment events, celebrity encounters, and hospital visits. However, staff struggle to locate specific images needed for donor communications, impact stories, and event marketing due to inconsistent naming conventions, staff turnover, and the limitations of traditional folder browsing. The challenge is intensified by the fact that images arrive from various photographers over decades, leaving new staff entirely unaware of event details from the past. Furthermore, child identities and highly confidential medical conditions must remain secure and ethically protected, meaning staff require a retrieval solution that can index natural-language scene descriptions and celebrity tags without incorrectly inferring or exposing sensitive medical diagnoses from raw image pixels.

Research and Solution Development

This project follows the Design Science Research Methodology and the OSEMN framework to build an AI-driven Photo Repository and Hybrid Search System. The technical pipeline will be designed around a Photo Enrichment Pipeline that extracts metadata, generates computer vision captions, and isolates scene or location tags (such as stadiums or hospitals). Additionally, a Hybrid Search Module is being built to support structured filters—such as CRM medical categories, celebrity tags, or dreamer names—alongside free-text semantic search over multimodal image embeddings.

Impact and Outcomes

This project is in its early initialization stages; therefore, specific system deployments and quantitative search evaluation metrics remain limited as the foundational components are established. The final pipeline aims to eliminate manual image tracking and overcome institutional knowledge gaps caused by organizational turnover. Once fully integrated with DCT’s database, this privacy-aware retrieval architecture will empower staff to instantly locate historical photos using intuitive natural-language queries, allowing them to efficiently pull visual evidence of dreams realized to drive fundraising and community engagement.

Student Researchers
Sri Chaithanya Chityala (Data Science)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing)

 

Project ID: #AI4DSSG-017
August 2025 – Current Pro-Bono / Unfunded Ongoing Research Project Thesis

Call Center Profiles: Matching Crisis Hotline Callers with the Right Help

Technologies: RTopic ModelingClustering

Community Partner: Here Tomorrow

Social Issue: Suicide Prevention Support

The first and most critical step in crisis support is quickly recognizing the unique emotional and situational needs of each caller, while recovery specialists track individual progress over a year-long care cycle to ensure lasting recovery. To enhance this vital work, this project examines how adult attachment styles and recurring conversational themes connect to specific risk factors. The goal is to build a structured framework that guides operators as they steer individuals toward the right care. Utilizing a foundation of over 8,000 records, the project creates specialized caller-response profiles to help Here Tomorrow staff confidently meet a caller’s most immediate needs.

Research and Solution Development

To deepen the understanding of each call, the investigation looked closely at the role adult attachment styles play during a crisis. Topic modeling served as a valuable tool to analyze written narrative notes and pull out the most common conversational themes. Finally, by bringing together these discussion topics, risk and protective factors, and call frequencies, clustering algorithms successfully grouped similar caller situations together to reveal clear patterns.

Impact and Outcomes

The analysis successfully revealed distinct caller profiles, shedding light on how different groups experience varying levels of risk and protective factors. Staff can now connect these profiles to specific intervention strategies, ensuring that every person who reaches out receives the exact care and support best suited for their unique situation.

Student Researchers
Brandon Lucas (Psychology)
DSSG Experts: Dr. Dan Richard (Psychology)

 

Project ID: #AI4DSSG-016
January 2026 – Current Funded Project Ongoing Research Project Stipended Position

Evaluating Trauma-Responsive Interventions: Survey Data Analysis and Pattern Extraction

Technologies: Statistical AnalysisSurvey Feedback AnalysisExploratory Data Review

Community Partner: Hope & Healing JAX

Social Issue: Trauma-Informed Care

Hope & Healing JAX collects survey responses from participants during their trauma-informed care training sessions in organizations and schools. However, this raw feedback remains unanalyzed, making it difficult for the organization to identify clear impact patterns or make sense of the training results. To better evaluate their programs and provide data-driven recommendations, they require a research-based analysis of their survey data to uncover how these sessions influence participant mindsets and caregiving strategies.

Research and Solution Development

The upcoming work focuses on setting up a simple workflow to clean, process, and analyze the collected training survey responses. Detailed technical specifications, data pipelines, and specific software tools will be shared later when the research design is fully finalized.

Impact and Outcomes

This project is currently in progress, and the primary goal is to establish a foundational review of the survey responses. Specific data patterns, final outcomes, and strategic recommendations will be added to this section at a later date once the data analysis is complete and the final results are generated.

Student Researchers
Faith Hussey (Sociology)
DSSG Experts: Dr. Dan Richard (Psychology)

 

Project ID: #AI4DSSG-015
January 2025 – Current Funded Project Ongoing Data Science Solution Stipended Position Thesis

Historical Analysis of Duval County Population and Housing Units at the Census Block Level as Indicators of Migration Patterns

Technologies: Geographic Relationship CrosswalksTIGER/Line ShapefilesUnsupervised ClusteringSpatial Hotspot AnalysisGeospatial Dashboard VisualizationsCensus Block DataPythonTableau

Community Partner: Local Initiative Support Corporation (LISC) Jacksonville

Social Issue: Demographic Shifts and Neighborhood Stabilization

LISC Jacksonville lacks a multi-decade, geographically comparable framework to track localized neighborhood evolution and demographic shifts across Duval County, leaving urban planners and community advocates without precise data on residential mobility. Evaluating longitudinal population and housing dynamics is severely hindered because U.S. Census Bureau block boundaries change every decade, causing complex one-to-many and many-to-many geographic boundary transformations. Without a stable, cross-decade spatial model, vital migration patterns—such as changing racial compositions, shifts in age structures, and localized housing fluctuations—remain obscured by shifting geographic baselines, preventing a clear understanding of which communities have experienced the most significant transformations.

Research and Solution Development

We are developing a reproducible longitudinal spatial analysis framework utilizing Decennial Census data from 2000, 2010, and 2020. To resolve the challenge of shifting boundaries, the methodology implements Census Bureau geographic relationship crosswalk files and a custom computational workflow to generate geographically stable longitudinal block groups referenced against TIGER/Line Shapefiles. The analytical pipeline filters out uninhabited land types, calculates inter-decadal temporal differentials, and applies unsupervised clustering, trajectory classification, and spatial hotspot analyses to identify areas experiencing migration-driven change. To make these complex spatiotemporal dynamics accessible, the solution incorporates an interactive geospatial dashboard mapping population and housing trends by race, age cohort, and spatial extent.

Impact and Outcomes

As an initiative currently in progress, this research successfully defines a robust, methodological framework and analytical design for tracking micro-level neighborhood evolution over a twenty-year span. By isolating migration flows from natural population increases, the completed pipeline architecture provides a reproducible model for handling boundary variations across multiple decades. The anticipated outcomes of this spatiotemporal model will equip LISC Jacksonville with data-driven indicators of residential mobility and housing evolution, establishing a foundational baseline to guide strategic community investment and neighborhood stabilization policies.

Student Researchers
Kushul Reddy Palakala (Computer Science) Laxman Reddy Adulla (Data Science) Rachel Reed (Sociology) Arsen Hoxha (Data Science) Mahmoud Elbatouty (Intelligent Systems)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing), Dr. Dan Richard (Psychology)

 

Project ID: #AI4DSSG-014
January 2026 – Current Funded Project Ongoing Data Science Solution Stipended Position Research Experience

Building a Data Pipeline for the Automation of Processing and Extraction of Relevant Data from City Budget Documents

Technologies: ClaudeNotebookLMGrokTableau PublicMulti-LLM Extraction PipelineR

Community Partner: Florida Nonprofit Alliance

Social Issue: City-Funded Nonprofit Fiscal Resilience

The Florida Nonprofit Alliance (FNA) lacks a standardized, scalable mechanism to track and compare municipal funding levels across local governments, leaving them unable to determine the exact percentage of city budgets allocated to community organizations. While municipal entities publish exhaustive city budget documents, these public records are notoriously opaque, dense, and formatted differently from one city to the next. The FNA requires a data-driven framework to isolate specific non-profit funding line items and discover macro trends in municipal support across different jurisdictions. Because manually parsing thousands of pages of municipal finances is resource-intensive and prone to data silos, an automated data science intervention is needed to efficiently extract and analyze localized funding configurations.

Research and Solution Development

We are designing and implementing a standardized data engineering pipeline to automate information extraction from broad, multi-city financial archives. To navigate tight resource constraints, the research team is focusing on a representative cohort of four major Florida cities: Jacksonville, Orlando, Tampa, and Miami. The methodology engineered a multi-LLM extraction architecture—initially testing Grok before shifting to Claude for deep textual parsing, while utilizing NotebookLM to capture isolated data segments for Jacksonville and Tampa. This automated extraction layout was verified under systematic human oversight and structural correction from FNA representatives, with the final parsed outputs loaded directly into Tableau Dashboard to enable interactive, comparative cross-city dashboard mapping.

Impact and Outcomes

The implementation of the automated parsing pipeline successfully transformed dense, heterogeneous municipal financial registries into clean, structured data streams, allowing for the rapid retrieval of targeted non-profit budget line items. By extracting and evaluating historical data configurations across the four major metropolitan regions, the platform successfully provided the Florida Nonprofit Alliance with unparalleled insights into city-level allocation patterns and municipal support structures. Ultimately, this foundational research successfully builds the baseline pipeline architecture needed to scale municipal document processing, allowing for significantly faster, automated, and more efficient budget trend analysis across a much wider array of Florida cities in the future.

Student Researchers
Zain Malik (Data Science) Indiana Ludwig (Data Science) Bryce Bentley (Psychology)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing), Dr. Dan Richard (Psychology)

 

Project ID: #AI4DSSG-013
January 2026 – Current Pro-Bono / Unfunded Ongoing AI Innovations Research Experience

Machine Learning Pipeline for Data Extraction and Analytics from Scanned PDF Forms for MMA-based Youth Program

Technologies: Optical Character Recognition (OCR)Layout-Aware Machine Learning ModelsDeep Learning Entity Recognition ModelsIntelligent Field ClassificationCSV ExportPython

Community Partner: Team Nitro MMA

Social Issue: Youth Engagement and Leadership Development

Paper-based data collection methods frequently hinder the digital transformation of nonprofit organizations like Team Nitro MMA, a youth mentorship program that gathers experience data from students and parents through paper questionnaires. These physical documents are digitized only after scanning into static PDF files, creating an operational bottleneck that stalls timely insight generation, program refinement, and the reporting required by granting agencies. Manual data entry is not a feasible solution due to its time-consuming, inaccurate, and unscalable nature as data volumes increase, while traditional rule-based parsing methods struggle with the inherent noise of scanned documents like handwriting and non-standard layouts. Consequently, Team Nitro requires an automated data science intervention to seamlessly handle both the layout-aware extraction of this unstructured text and the deep data analysis of the extracted results, as they cannot currently perform the predictive modeling, participant outreach assessments, or trend visualizations necessary for data-driven governance.

Research and Solution Development

We are designing and implementing an end-to-end Machine Learning (ML) and Deep Learning (DL)-powered document processing pipeline. The methodological solution shifts away from rigid, template-reliant parsing by integrating context-aware Optical Character Recognition (OCR) models to process document images into raw text, alongside layout-sensitive models and deep learning entity recognition models to automatically interpret document structures. By combining layout-aware information extraction with intelligent field classification, the pipeline accurately maps heterogeneous text and inconsistent markings to a predefined data schema. The extracted data is then systematically cleaned, validated, and normalized to ensure high-fidelity data curation.

Impact and Outcomes

The implementation of this document AI system successfully automates the transition from manual digitization to automated extraction, converting noisy scanned PDF questionnaires into clean, structured, and machine-readable CSV datasets. By eliminating the manual data entry bottleneck, the pipeline enables significantly faster analytics turnaround times and more frequent program evaluations for Team Nitro. Unlocking these high-fidelity datasets directly facilitates sophisticated downstream capabilities—including descriptive analytics, interactive dashboards, predictive modeling, and outcome simulations. Ultimately, this structured framework will empower Team Nitro with data-driven governance to optimize participant outreach, evaluate program design, and demonstrate their community impact to securing grants.

Student Researchers
Sandeep Kumar Bontha (Data Science) Rohith Varakoti Srinivas Reddy (Data Engineering)
DSSG Experts: Dr. Dan Richard (Psychology), Dr. Karthikeyan Umapathy (Computing)

 

Project ID: #AI4DSSG-012
August 2025 – Current Pro-Bono / Unfunded Ongoing AI Innovations Research Experience

Food4Medicine: End-to-End Machine Learning and Automated Fulfillment Pipeline for Personalized Weekly Grocery Recommendation and Recipe Generation

Technologies: RandomForestClassifierTF-IDF VectorizerNearestNeighborsMinMaxScalerGoogle OR-ToolsT5 Recipe GenerationGradioJSONPythonHugging Face

Community Partner: Epic-Cure

Social Issue: Food Insecurity and Chronic Disease Management

Medically vulnerable patients enrolled in community food bank programs face a critical gap between receiving medically tailored food recommendations and real-world grocery fulfillment. While standard recommendation workflows can identify ideal foods for chronic conditions like diabetes, heart disease, or stroke, they do not account for immediate food bank inventory limitations, allergen safety, or the logistical strain of physically gathering items across a store or food reserve. Furthermore, patients often lack guidance on how to transform a weekly bag of raw ingredients into structured, health-compliant daily meals. Epic-Cure requires an automated system to seamlessly handle clinical risk stratification, inventory-aware substitutions, physical route optimization, and personalized meal planning because community food initiatives struggle to deliver practical, scalable, and safe nutritional interventions manually.

Research and Solution Development

To bridge this gap from clinical data to physical delivery, a full-stack artificial intelligence and data engineering system was developed over a multi-semester research initiative. The machine learning pipeline processes data through a series of specialized models: a MinMaxScaler normalizes clinical biomarkers from electronic health profiles, a RandomForestClassifier performs tri-class patient risk stratification, a TF-IDF Vectorizer with cosine similarity handles semantic food retrieval, and a NearestNeighbors model executes nutritional proximity ranking. To translate these recommendations into real-world fulfillment, the pipeline utilizes USDA FoodData Central nutrition data and disease-specific nutrient weightings to select available substitutes when inventory is low, while Google OR-Tools calculates an optimized shopping route through the store. Finally, the system integrates a T5 recipe generation model (flax-community/t5-recipe-generation) checked against ADA and NHLBI DASH dietary guidelines to output a 7-day meal plan, operating through a web interface using Gradio.

Impact and Outcomes

The integrated pipeline achieved complete automation of the medically tailored grocery workflow, executing a synthetic 500-patient cohort simulation in just 7.64 seconds (15.3 ms per patient) with 100% verified allergen safety. The system successfully mapped a clinical distribution of 8.8% high-risk, 24.0% moderate-risk, and 67.2% low-risk patients while accurately surfacing 1,064 total clinical alerts and generating structured JSON outputs compatible with electronic health records (EHR) and logistics systems. About 30% of parent survey responses across previous iterations validated that this style of support maintains creative capabilities. Ultimately, the system empowers food bank staff to upload inventory files and generate a comprehensive 7-day meal plan with automated nutrition analysis in 10 to 15 minutes, establishing a scalable model for evidence-based chronic disease management.

Student Researchers
Cibi Siddarth (Computer Science) Manikanta Chowdary Musunuru (Data Science) Nagaraju Gulakaram (Data Science)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing)

 

Project ID: #AI4DSSG-011
August 2025 – Current Pro-Bono / Unfunded Ongoing AI Innovations Research Experience

A Multimodal Deep Learning Pipeline for Person Identification, Action Recognition, and Context-Aware Behavioral Analysis from Unstructured Video Data

Technologies: MTCNNFaceNetFER-CNNXCLIPWhisperLibrosaAgglomerative ClusteringLarge Language Models (LLMs)PythonHugging Face

Community Partner: The Performerance Academy

Social Issue: Arts-Based and Trauma-Informed Positive Youth Development

The rapid proliferation of unstructured video archives within social service programs, such as The Performance Academy (TPA) performance records, presents a major technical challenge for qualitative assessment. Valuable student data—including artistic expressions like dancing, modeling, spoken-word delivery, and deep personal reflections—remains locked inside raw multimedia formats that are too complex and computationally expensive for manual analysis. Traditional action recognition architectures and classification pipelines require massive training datasets and extensive preprocessing, creating a significant barrier to extracting critical behavioral, emotional, and context-aware insights. To overcome these data limitations and support evidence-based evaluation, TPA requires a scalable, automated pipeline capable of transforming unstructured performance videos into structured documentation of student reflections and behavioral statistics.

Research and Solution Development

To address these video processing and resource constraints, a comprehensive, memory-efficient multimodal analytics pipeline was engineered to process long video files via a specialized frame-sampling and chunking strategy. The pipeline integrates a suite of deep learning techniques: Multi-task Cascaded Convolutional Networks (MTCNN) for face detection, FaceNet embeddings with Agglomerative Hierarchical Clustering for cross-video identity tracking, and a CNN-based Facial Expression Recognition (FER) model for emotion tracking. Human actions are mapped using XCLIP—a transformer-based vision-language model that utilizes textual prompts to interpret expressive creative movements—while audio signals are decoded using Whisper for text transcription and Librosa for rhythm detection. Large Language Models (LLMs) were then benchmarked to evaluate, identify, and extract specific student reflections from the generated transcripts, outputting structured data files complete with raw reflection content, video titles, and exact timestamps.

Impact and Outcomes

The implementation of this multimodal framework successfully automated the extraction of structured, person-wise appearance statistics and chunk-level behavioral summaries from raw youth performance archives. By isolating key thematic contexts such as teaching, poetry, and rhythmic engagement without requiring manual annotations, the system provides a robust model for behavioral research and education analytics. Most importantly, the resulting AI-extracted documentation preserves tangible student reflections, equipping TPA with the concrete, verifiable data necessary to systematically assess its programmatic impact and secure critical public or private funding to expand its performance arts services for adolescent youth in Jacksonville.

Student Researchers
Laxman Reddy Adulla (Data Science) Jaswanth Golla (Data Science) Indiana Ludwig (Data Science)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing)

 

Project ID: #AI4DSSG-010
August 2025 – April 2026 Pro-Bono / Unfunded Completed AI Innovations Research Experience

Audio-to-Knowledge Modeling for Structured Insights from Disengaged Youth Interviews

Technologies: NVIDIA NeMoParakeet ASRGemini APIBERTopic ModelingReactffmpegPythonHugging Face

Community Partner: The Performerance Academy

Social Issue: Work Readiness

Disengaged young adults encounter significant barriers to career readiness due to interruptions in education, limited workforce exposure, and a lack of structured guidance. While The Performance Academy (TPA), through The Opportunity Project (TOP), conducts intake interviews to understand participants’ backgrounds, challenges, and goals, these recordings are currently analyzed manually. This manual workflow restricts the ability to extract broader thematic patterns across participants, limiting data-driven program design and personal coaching scalability for vulnerable youth.

Research and Solution Development

An end-to-end speech processing and Natural Language Processing (NLP) pipeline was engineered to automate qualitative analysis. The system uses ffmpeg for audio preprocessing, NVIDIA NeMo’s ClusteringDiarizer for speaker diarization, the Parakeet ASR model for segment-level transcription, and the Gemini API to format transcripts into structured question-answer pairs and concise summaries. To identify overarching themes across a processed dataset of 19 intake interviews, BERTopic modeling was integrated into the pipeline, alongside a React user interface that enables staff to upload audio, track processing history, and download structured reports in PDF format.

Impact and Outcomes

The automated pipeline successfully transformed conversational interview audio into decision-ready data, extracting 296 participant-focused text segments and mapping 11 major thematic groups. These discovered themes uncovered vital participant contexts, including education pathways and GED completion, job readiness, mental health challenges, career aspirations in trades and creative fields, transportation barriers, family context, and criminal justice history. By replacing manual analysis with this scalable framework, the solution provides TPA staff with the actionable insights necessary to quickly identify common student needs, enabling data-driven program design and more personalized workforce coaching.

Student Researchers
Anjali Prakash (Data Science) Adhert Johnson (Data Science)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing)

 

Project ID: #AI4DSSG-009
August 2025 – June 2026 Funded Project Completed Research Project Stipended Position

Igniting Smarts Beyond the Traditional Classroom: An Assessment of Student Impact and Program Delivery Across CAP's Arts Ignite and Art Smarts Initiatives

Technologies: TableauPythonSurvey ResponseStatistical Analysis

Community Partner: Cathedral Arts Project

Social Issue: Arts Education

Across Clay, Duval, Nassau, and St. Johns counties, students in charter schools, home-school networks, military communities, and residential crisis care face a profound lack of specialized arts education outside the traditional public school framework. To bridge this critical resource gap, the Cathedral Arts Project (CAP) partnered with the Florida Department of Education (FLDOE) to launch a bold initiative utilizing its core Arts Ignite afterschool enrichment and Art Smarts academic integration models, which were recently extended to military families at Mayport Naval Base and local Blue Star schools. <br> <br> While CAP is actively addressing this regional gap through these diverse delivery models, what is the precise student impact of these new interventions? Determining this is critical to understanding exactly how this specialized instruction advances the academic, emotional, social, and therapeutic development of these unique student populations, driving CAP's need for a comprehensive program assessment and impact analysis.

Research and Solution Development

To optimize program delivery, a multi-phase trend analysis was conducted using student and family survey data to evaluate student perceptions, behaviors, and experiences across various arts disciplines. Natural Language Toolkit sentiment analysis was applied to open-ended caregiver feedback to gauge the overarching tone of program reception, revealing a highly favorable compound score of 0.713 with a 53.7% positive and 0.4% negative sentiment split. Based on research showing that students heavily enjoyed lessons incorporating math, reading, and science experiments (such as states of matter, astronomy, and erosion), the primary solution recommendation is to systematically expand the inclusion of these core academic subjects into the arts curricula to maximize multi-disciplinary engagement.

Impact and Outcomes

The evaluation demonstrates that participation in CAP programming directly yields strong improvements in student mental focus, frustration management, and academic persistence. Over 50% of surveyed parents reported measurable gains in their child's creative problem-solving skills, while the rate of observed positive gains in student social skills rose over time by 2.6% to reach 72.6%. Furthermore, the programming successfully fostered deeper familial connections, with an average of 43.6% of caregivers reporting enhanced child-parent interactions and 56.6% noting that their children returned home explicitly excited to share what they learned.

Student Researchers
Ella Luedeke (Computer Science)
DSSG Experts: Dr. Dan Richard (Psychology), Dr. Karthikeyan Umapathy (Computing)

 

Project ID: #AI4DSSG-008
August 2025 – May 2026 Funded Project Completed Data Science Solution Stipended Position

State of Jax Initiative Dashboards

Technologies: SQLTableaumySidewalk

Community Partner: City of Jacksonville

Social Issue: Benchmarking Cities

The State of Jax initiative benchmarks Jacksonville's progress and socio-economic conditions by comparing Jacksonville’s performance directly against its peer cities.

Research and Solution Development

The project extracts community data from the mySidewalk platform and develops 30 public-facing interactive Tableau dashboards to visualize multi-sector indicators across Jacksonville’s neighborhoods and council districts.

Impact and Outcomes

The resulting dashboards provide clear, side-by-side regional comparisons that highlight exactly where Jacksonville excels or lags behind its peer cities based on standardized metrics.

Student Researchers
Sneha Kattari (Data Science)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing), Dr. Dan Richard (Psychology)

 

Project ID: #AI4DSSG-007
January 2025 – December 2025 Pro-Bono / Unfunded Completed Data Science Solution Research Experience

Comparative Urban Analytics: Application of ISO 37120 for Benchmarking Jacksonville against Major Midsized U.S. Metropolitan Areas

Technologies: SQLTableaumySidewalkPython

Community Partner: City of Jacksonville

Social Issue: Benchmarking Cities

The City of Jacksonville, Florida lacks a centralized and standardized method to evaluate its social, economic, environmental, and infrastructural performance against other major U.S. metropolitan areas. Researchers and policymakers are forced to navigate heterogeneous, raw public data sources across fragmented domains like health services, housing stability, economic inequality, food security, and public safety. Without a comparative model, crucial municipal challenges remain unaddressed and obscured, such as persistent low drinking water compliance rates, a lagging life expectancy of 73.1 years, and clear deficits in digital connectivity and transportation efficiency. To resolve this, a scalable performance tracking mechanism is required to process raw civic data into standardized indicators that can support evidence-based decision-making and guide urban planning, investment, and policy development.

Research and Solution Development

To address this municipal benchmarking gap, this project successfully built a data processing pipeline based on the OSEMN framework that integrates heterogeneous public data sources, including the U.S. Census Bureau, EPA, BLS APIs, the CMS National Provider Index, Feeding America, Zillow, and mySidewalk. The raw datasets were systematically cleaned, standardized, and utilized to construct a series of interactive, comparative dashboards in Tableau. The methodological solution benchmarks Jacksonville against a diverse cohort of 13 peer cities by calculating Key Performance Indicators (KPIs) defined by the ISO 37120 standard for sustainable cities and communities across themes like Environment, Urban Planning, and Water Quality. This research establishes a reproducible model for municipal performance tracking that lays the groundwork for future GUI finalization, greater geographic granularity at the census tract level, and expansion into ISO 37122 (Smart Cities) and ISO 37123 (Resilient Cities) standards.

Impact and Outcomes

The comparative urban analytics dashboard successfully generated crucial performance insights, revealing that Jacksonville performs competitively in unemployment outcomes and demonstrates remarkable housing stability with a 2023 average home value of approximately $361,000, positioning it as a mid-tier growth market. Environmental metrics also exposed measurable progress in regional greenhouse gas emissions reduction. However, the data simultaneously isolated distinct local vulnerabilities, proving that Jacksonville faces challenges in digital connectivity (broadband internet access), transportation efficiency (average commute time), and a persistent low drinking water compliance rate. Furthermore, the public health indicators revealed a major deficit, showing that Jacksonville’s life expectancy of 73.1 years lags behind its top-tier peer cities.

Student Researchers
Jason Gardner (Data Science) Gnana Karthik Yalamanchi (Data Science) Sai Rithwik Nooguri (Data Science)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing)

 

Project ID: #AI4DSSG-006
August 2024 – June 2025 Funded Project Completed Research Project Stipended Position

Arts for All initiative: Evaluating the Impact of Specialized Arts Education on Student Development

Technologies: TableauPythonSurvey ResponseStatistical Analysis

Community Partner: Cathedral Arts Project

Social Issue: Arts Education

Across Clay, Duval, Nassau, and St. Johns counties, students in charter schools, home-school networks, military communities, and residential crisis care face a profound lack of specialized arts education outside the traditional public school framework. While the Cathedral Arts Project is actively addressing this regional gap by delivering tailored after-school, home-school, and arts integration programming, what is the precise student impact of these new interventions? Determining this is critical to understanding exactly how this specialized instruction advances the academic, emotional, social, and therapeutic development of these unique student populations.

Research and Solution Development

To understand the exact multi-week trajectory of participant experiences, we structured our research to analyze multi-period survey trends collected across the Fall and Winter survey cycles. This research focused on evaluating student outcomes across multi-week survey periods to identify exactly how the arts programming influences behavioral and academic persistence. By examining these trends over time, we developed tailored solutions and actionable recommendations to help instructors manage high-need classrooms and provide families with at-home reinforcement materials.

Impact and Outcomes

Our findings show that the programming successfully drives positive student outcomes in communication, persistence, and emotional regulation, with parents noting fewer instances of frustration when students tackle difficult tasks. By the winter survey period, approximately 30% of parents reported measurable improvements in their child's creative problem-solving, alongside significant increases in social interaction and sharing within the home. While a few narrative responses noted localized classroom management challenges and a minor seasonal dip in theater confidence, overall ratings for enjoyment and artistic competence remained exceptionally high.

Student Researchers
Bryce Bentley (Psychology)
DSSG Experts: Dr. Dan Richard (Psychology), Dr. Karthikeyan Umapathy (Computing)

 

Project ID: #AI4DSSG-005
August 2024 – June 2025 Pro-Bono / Unfunded Completed Data Science Solution Research Experience

Assessing the Impact of Accreditation on the Fiscal Health of Nonprofits

Technologies: SQLTableauPythonIRS 990 DataLUNA model

Community Partner: Nonprofits First

Social Issue: Fiscal Health of Nonprofits

Nonprofits First provides accreditation services and oversight to drive administrative excellence, financial accountability, and standard compliance across the sector. They want to analyze their robust dataset of over 72 accredited organizations to identify the key operational factors and relationships that ultimately boost nonprofit credibility, fiscal health, and funding accessibility.

Research and Solution Development

We developed an interactive analytics dashboard equipped with dynamic filters for accreditation status, geographic area, and asset-size comparison groups. The platform instantly visualizes critical near-term liquidity and long-term stability metrics—including LUNA, debt-to-asset, and administrative cost ratios—for individual or grouped nonprofits across multiple tax years.

Impact and Outcomes

Analysis and dashboard visuals validates the effectiveness of the accreditation process by proving that Nonprofits First's rigorous standards lead to superior long-term fiscal health and organizational sustainability. By transforming complex auditing metrics into clear, year-over-year trends, the dashboard provides the verified data evidence needed to justify strict oversight requirements while giving stakeholders and prospective donors undeniable proof of the credential's value.

Student Researchers
Jonathan Beckham (Data Science)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing), Dr. Dan Richard (Psychology)

 

Project ID: #AI4DSSG-004
August 2024 – April 2025 Pro-Bono / Unfunded Completed Data Science Solution Research Experience

Drawing Insights on Florida’s Voter Engagement Using Machine Learning Approaches

Technologies: TableauPythonFlorida Voter DataCensus DataPredictive ModelingClustering

Community Partner: League of Women Voters of Florida

Social Issue: Voter Engagement

Non-partisan organizations like the League of Women Voters of Florida (LWVFL) struggle to efficiently allocate resources and identify which specific communities need the most encouragement to vote. Public registration and Census data from 2012 to 2020 were analyzed to identify hidden voting patterns and address this critical gap in targeted outreach. This baseline analysis allows LWVFL to pinpoint exactly where historical turnout drops occur without compromising individual voter privacy.

Research and Solution Development

Predictive modeling involved training Logistic Regression, Random Forest, and Boosted Trees algorithms to predict individual voter turnout in the 2020 federal general election based on features like age, gender, ethnicity, and zip code. Additionally, a K-Means clustering algorithm was fine-tuned using 2018 voting data to group zip codes with similar behaviors. This clustering model was then applied to predict voter groups for the 2020 election.

Impact and Outcomes

Tangible deliverables include an optimized K-Means clustering algorithm mapping zip code behaviors and verified turnout prediction models for individual voters. The final outcome delivers a new plan of action designed to help non-partisan organizations optimize their outreach efforts. It provides a strategic guide detailing which specific voter clusters and individuals to target to most effectively promote voter turnout in Florida for future elections.

Student Researchers
Mahmoud Elbatouty (Data Science)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing)

 

Project ID: #AI4DSSG-003
August 2023 – May 2024 Pro-Bono / Unfunded Completed Data Science Solution Thesis Independent Study

A clustering-based approach to analyze impacts of digital divide on undercounting issues in census in Florida

Technologies: PythonClusteringTableauCensus DataGIS Mapping

Community Partner: Florida Philanthropic Network, Inc

Social Issue: Census Undercount and Digital Divide

The online-heavy census framework risks creating a massive undercount across Florida because severe gaps in internet access and technology leave countless residents entirely invisible. Florida Philanthropic Network (FPN) wanted to know the exact counties and zip codes most vulnerable to this digital divide gap. FPN wanted a clear, data-driven map pinpointing these high-risk geographic areas so their network of nonprofits and funders can understand precisely where community resources and advocacy are missing ahead of the 2030.

Research and Solution Development

We developed an unsupervised machine learning solution using K-Means and Hierarchical clustering algorithms to analyze county and zip-code level data across Florida. This solution groups regions based on critical features like broadband subscription rates, computer ownership, poverty levels, and historical census return rates. By translating these complex clusters into an interactive GIS geographic dashboard, we provide FPN with a precise visual map of the exact zip codes where low connectivity directly triggers high undercount risks.

Impact and Outcomes

This clustering solution provides FPN with an objective, data-backed blueprint that identifies exactly which specific zip codes and demographics face the highest technological barriers. By knowing precisely where the digital divide overlaps with undercount risks, the network can eliminate guesswork and base its strategic planning on verified geographic insights. Ultimately, this specific data tool ensures that FPN has the empirical clarity needed to target its regional advocacy and resource alignment exactly where the technology gaps are most severe.

Student Researchers
Partha Protim Datta (Data Science) Tushya Vemuri (Data Engineering)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing), Dr. Dan Richard (Psychology), Dr. Xudong Liu (Machine Learning), Dr. Sandeep R. Reddivari (Software Engineering)

 

Project ID: #AI4DSSG-002
August 2023 – April 2024 Pro-Bono / Unfunded Completed Research Project Research Experience

Investigating Google Search Trends Patterns in Relevance to Disaster Events

Technologies: PythonTableauGoogle TrendsGoogle Analytics

Community Partner: Disaster Central

Social Issue: Community Disaster Resilience

Disaster Central wants to enhance community disaster resilience by analyzing digital behavior and public information needs surrounding major Florida hurricanes in 2022 and 2023.

Research and Solution Development

The project models search interest trends and digital resource navigation by integrating Google Trends data and Google Analytics metrics captured during Hurricanes Ian and Idalia.

Impact and Outcomes

The findings provide actionable data insights to help Disaster Central optimize real-time communication strategies, anticipate critical resource demands, and proactively deploy digital aid to Florida communities during future climate crises.

Student Researchers
Everett Chambler (Data Science)
DSSG Experts: Dr. Karthikeyan Umapathy (Computing), Dr. Dan Richard (Psychology)

 

Project ID: #AI4DSSG-001
January 2021 – August 2022 Pro-Bono / Unfunded Completed Research Project Research Experience

Building Generational Wealth: Intergenerational Impacts of the HabiJax Affordable Housing Program

Technologies: Qualitative ResearchTableau

Community Partner: First Coast Habitat For Humanity

Social Issue: Affordable Housing

HabiJax wanted to evaluate the long-term socio-economic impacts of its affordable housing services by understanding how transitioning into affordable homeownership affects community stability, financial health, and family well-being. Specifically, the project sought to understand how affordable homeownership fosters intergenerational wealth building to directly address and break the cycle of poverty.

Research and Solution Development

The study analyzed qualitative interviews with eight HabiJax homeowners using the Consensual Qualitative Research (CQR) method to map the homeowner journey across ten distinct thematic domains. The outcomes track participant experiences from their housing circumstances and motivations before buying a home, through their short- and long-term milestones, including emotional perks and the specific life crises where a HabiJax home provided a safety net. The findings also capture the proactive actions participants took to help themselves and compile direct homeowner advice to guide where HabiJax can continue to optimize its assistance.

Impact and Outcomes

The insights from these ten domains empower HabiJax to quantify the holistic, multi-generational value of affordable housing, shifting the narrative from basic shelter to long-term community stability and poverty alleviation. By measuring both immediate benefits and long-term socio-economic shifts for individuals and families, the project provides data-driven evidence to refine future housing services, maximize community well-being, and advocate for strategic affordable housing policies.

Student Researchers
Athlene Jones (Data Science) Kelly (Interviewer) Jacob (Interviewer) Darlene (Interviewer) Hazel (Qualitative Researcher) Christina (Qualitative Researcher)
DSSG Experts: Dr. Tes Tuason (Public Health), Dr. Dan Richard (Psychology), Dr. Karthikeyan Umapathy (Computing)