Proving the Future of Institutional AI: The DHBW Lörrach Partnership

In a groundbreaking initiative that serves as a crucial proof of concept for InstitutionalAI, Devolved AI and DHBW Lörrach have partnered to develop a specialized AI system that is simultaneously advancing educational AI capabilities and validating next-generation institutional AI deployment methods. This project represents more than just an educational experiment – it's a carefully designed prototype that demonstrates how institutions can develop, train, and maintain their own specialized AI systems.

The Dual Purpose: Education and Validation

The DHBW Lörrach project operates on two distinct levels. First, it provides students with hands-on experience in AI development. Second, and perhaps more significantly, it serves as a proving ground for InstitutionalAI's revolutionary approach to deploying specialized AI systems within organizations.

What makes this project particularly valuable as a proof of concept is that students are manually implementing components that will eventually be automated and streamlined in the InstitutionalAI platform. While students work directly with raw data preprocessing, scoring systems, and model fine-tuning, they're simultaneously validating the processes that will be packaged into InstitutionalAI's intuitive frontend interface.

Current Implementation and Future Platform

Today's Manual Development

Under the weekly supervision of ML developer Jay Kim, 2-3 students are deep in the trenches of AI system development. They're working with:

• Raw data preprocessing pipelines, manually handling data transformation and preparation
• Direct model fine-tuning processes, adjusting parameters and optimizing performance
• Custom-built scoring systems to evaluate model improvements
• The Athena 2 codebase, which serves as the foundation for their development

Tomorrow's Streamlined Platform

The insights gained from this manual implementation are directly informing the development of InstitutionalAI's commercial platform, which will offer:

• A sophisticated frontend interface that unifies all components
• Pre-fine-tuned models ready for institutional deployment
• Automated preprocessing pipelines that handle data preparation without technical intervention
• Integrated scoring systems that automatically evaluate and optimize performance

This contrast between manual implementation and future automation is precisely what makes the project valuable as a proof of concept. Every challenge the students face helps refine the automated solutions that will be built into InstitutionalAI.

Decentralized Training in Action

The project employs an innovative multi-processor training methodology that demonstrates how specialized AI systems can be optimized for institutional use. The current implementation utilizes:

• 1x A100 GPU
• 2x L40 GPU
• 4x GeForce GPU
• 3x A10 GPU

These processors operate within a sophisticated competitive evaluation framework. For example, when the A100 achieves a 20-point improvement while the L40 contributes 1 point, the system automatically weights the model updates proportionally (20/21 from A100, 1/21 from L40). Lower-performing processors are systematically replaced by better alternatives from a waiting pool, ensuring optimal resource utilization.

Building the Foundation for Future Implementation

The project's current development phase focuses on several key areas:

1. Data Integration and Training
   • Students actively feed institutional data into the system
   • Custom training workflows are being developed for educational contexts
   • Weekly iterations improve the model's understanding of institutional specifics
2. System Architecture Development
   • Building on the Athena 2 code base
   • Implementing specialized training protocols
   • Developing scoring and evaluation systems

Expected Outcomes and Deliverables

This proof of concept will yield several crucial deliverables:

1. A Specialized Educational AI Model
   • Trained specifically on DHBW Lörrach's institutional data
   • Capable of supporting various educational functions
   • Designed for continuous improvement year over year
2. Validated Implementation Framework
   • Proven methodologies for institutional AI deployment
   • Documented best practices for system optimization
   • Clear pathways from manual to automated implementation
3. Academic Documentation
   • Comprehensive research findings
   • Detailed methodology documentation
   • Performance analysis and results
4. Educational Impact
   • Practical AI development experience for students
   • Reusable learning resources for future classes
   • Foundation for expanded AI education programs

Looking Forward

As the project moves toward completion, several key milestones lie ahead:

1. Refinement of processor performance evaluation metrics
2. Completion of initial training cycles
3. Documentation of findings for academic publication
4. Implementation of sustainable update mechanisms

More importantly, this project is providing crucial validation for InstitutionalAI's approach to deploying specialized AI systems. The manual work being done by students today is directly informing how these processes will be automated and streamlined in the final platform, ensuring that future institutions can benefit from specialized AI systems without requiring the same level of technical expertise.

The Bigger Picture

This initiative represents more than just a successful academic-industry partnership – it's a crucial step toward democratizing institutional AI deployment. By proving that specialized AI systems can be effectively developed and maintained within an institutional context, the project is paving the way for broader adoption of AI technologies across educational and organizational settings.

The distinction between the current manual implementation and the future streamlined platform is particularly significant. While students at DHBW Lörrach are gaining invaluable experience through hands-on development, their work is simultaneously validating the automated processes that will make institutional AI deployment accessible to organizations worldwide through the InstitutionalAI platform.

This project demonstrates not just how educational institutions can develop specialized AI systems, but how the future of institutional AI deployment will look. The results will benefit not only DHBW Lörrach but also provide a validated framework for institutional AI implementation worldwide.

‍