Based on the annual report, here are the significant results for the Sites & Stories Application and AI approaches that Suzanne Pierce and Texas Advanced Computing Center contributed to:
1. Deployed Scalable AI-Enabled Analytical Infrastructure
The Sites & Stories (S&S) application advanced from testing to full deployment as a reusable tool on Texas Advanced Computing Center Resources. The system now provides cross-scale applications of Natural Language Processing, topic modeling, and Large Language Model capabilities through a science gateway data portal that enables analyses at scale with HPC and GPU compute. LLM tools were successfully incorporated into the S&S application suite with open and reproducible access as of June 2024.
2. Validated Field-Ready Data Collection and Analysis Capabilities
The team completed 25 performance field tests in Nunapitchuk to evaluate the Sites & Stories data collection and ingest services using variations in hardware, software, and data catalog capabilities. These tests demonstrated the system's ability to function in remote Arctic environments while supporting both interview data collection and drone terrain data processing. The application successfully automated analytical approaches with AI to connect qualitative and quantitative knowledge in well-defined workflows.
3. Advanced Semantic Linking and Automated Knowledge Integration
The S&S application was enhanced with semantic linking mechanisms to automate identification of relevant data collections and models for locally identified stakeholder problems. The system now enables semantic linking of stakeholder interview information with digital terrain models, creating culturally-informed technical resources. This advancement allows for automated connection of community knowledge with scientific models and observational data.
4. Demonstrated Transferability and Scalability Across Contexts
The capabilities were successfully replicated and tested for transferability in the Rio Grande Valley of Texas, proving the system's applicability beyond Arctic contexts. The S&S application was deployed for broad use by researchers with training provided to large research teams and community partners. The system is now fully scalable, reusable, and easily modified across different settings and use cases.
5. Established Reusable Computational Cookbook Framework
AI-enabled analytical workflows are now available across multiple scales, from individual narrative analysis through Natural Language Processing computational cookbooks to topical analysis using BERTtopic cookbook, plus open-ended analysis tools using four Large Language Foundation Model options on TACC systems. These workflows provide mechanisms for linking lived experience information with observational data and physics-based models in convergent applications, with the full code base and reusable products available via Science Gateway applications and GitHub.
