Ecosystem Smart Niederbayern

Ready for a challenge?

Innovation Challenge for Students in Dingolfing/Landau

Ready for a challenge in the high-tech hub of Dingolfing-Landau? Join our innovation challenge and tackle real-world problems in the fields of lean, green and digital. Collaborate with an interdisciplinary team to develop practical and innovative solutions for companies, cities and regional partners. Expect exciting insights into the Ecosystem Smart Niederbayern, valuable networking opportunities and the chance to present your idea.

registration image

Register now!

When: 08. Okt. 2026 to 10. Okt. 2026
Where: Stadthalle Landau, Stadtgraben 3, 94405 Landau an der Isar

  • Due to limited capacity, the number of participants is limited. Secure your place and apply now!
  • You can find the descriptions of each use case below. Take a look to see which challenge fits your interests and skills.
Register by 13 September 2026

Not an IT-Nerd? No problem! Whatever you study – please apply. We are looking forward to having various perspectives on our use cases.

Simply complete the survey. Registration will end automatically once all places are filled.

Registration

What to expect at our Innovation Challenge

Use Cases

Motivation

"Smart Region starts with smart people!". Agent for reducing manual administrative work for managers.

Target

Automate the ordering process for initial supplies for new employees

Approach

  • Build a concept how to get important knowledge out of files for building an individual checklist. Also bild an Agent concept for doing the things automatic on the list.

Actors/Stakeholders

relevant Managers

Business Value

3 hours per admission

Distribution of Activities

  • 40% development and programming
  • 30% user experience design
  • 30% AI training and integration

Technical prerequisites

N8N (AWA) Instance, GitHub Copilot , Sharepoint space

Candidate profile

  • Bachelor or Master, N8N & GitHub Copilot knowledge

Motivation

No standardized BMW solution exists for automated GS91010-2 compliant DMC validation. Current checks are manual, error-prone, and time-consuming, leading to rework, delayed feedback, and potential line stoppages.

Target

Faster error detection, improved traceability, reduced downtime.

Approach

  • Capture DMC; Convert to structured plain text; automated rule-based validation; clear error feedback (position, format, content); Integrate into existing quality workflows; Use AI-supported development

Actors/Stakeholders

Thomas Kerscher, MO-383; Alexander Ottenbacher TS-424

Business Value

Lower manual effort in DMC validation; Faster complaint resolution; Improved traceability & compliance ; High scalability across BMW plants; Blueprint for AI-driven development use cases

DoD/Success Metrics / KPIs

Fully functional app for DMC scan, validation, and error explanation, reducing Line stoppages or issues caused by DMC Errors

Distribution of Activities

  • 45% Development & programming
  • 20% Rule logic & validation (GS91010-2)
  • 20% AI integration; 10% UX/UI design
  • 5% Testing & rollout

Technical prerequisites

No fixed technical constraints – solution-driven development approach

Candidate profile

  • Skills in software/app development; Understanding of data validation & rules logic; Experience with AI-assisted development; Background: Computer Science, Engineering, or similar; Optional: quality / production / traceability knowledge

Additional information

Strong scalability potential across BMW Group

Motivation

Warehouse staff in goods receipt currently switch constantly between physical handling of pallets and manual data entry on handheld scanners or desktop terminals. Every interruption costs time, increases error rates and slows down the entire incoming-goods process – especially in peak periods. Hands-free, voice-driven interaction with the warehouse management system removes this friction.

Target

Develop a voice-controlled mobile assistant that integrates with the SPG Warehouse Suite and lets warehouse staff perform goods-receipt postings (article, quantity, bin location, batch/serial) entirely hands-free, with the same data quality as today’s manual entry but at significantly higher throughput.

Approach

  • Define a constrained voice grammar for goods-receipt commands (article, quantity, bin, batch).
  • Build a Progressive Web App (PWA) for headset/smartphone use that captures audio via Azure AI Speech.
  • Use Azure OpenAI (GPT-4o) to convert spoken commands into structured JSON postings.
  • Connect the PWA via REST API to a mocked SPG Warehouse Suite backend (provided by SPG).
  • Implement voice-based confirmation, error correction and a real-time bin-occupancy view.

Actors/Stakeholders

Warehouse operator (primary user), warehouse supervisor, SPG Warehouse Suite product team, SPG development team, end customers of the SPG Warehouse Suite (logistics & wholesale companies in the DACH region).

Business Value

30–40% faster goods-receipt postings, lower error rate due to voice-based double confirmation, improved ergonomics and worker satisfaction, differentiating feature for the SPG Warehouse Suite in tenders against competing WMS products, reusable speech-interface pattern for further SPG products.

DoD/Success Metrics / KPIs

Working PWA demo on a smartphone with headset; minimum of 5 supported voice commands; speech-recognition accuracy ≥ 90% in a noisy demo environment; end-to-end posting latency < 3 seconds; complete demo of a 10-pallet goods receipt at the final pitch.

Distribution of Activities

  • 50% development and programming (PWA frontend, API integration, prompt engineering)
  • 25% AI training and integration (Azure Speech + Azure OpenAI)
  • 15% user experience design (voice-flow design, confirmation patterns)
  • 10% testing and demo preparation

Technical prerequisites

Azure AI Speech (Speech-to-Text, Text-to-Speech), Azure OpenAI Service (GPT-4o), Azure App Service for hosting the PWA, Azure API Management for the WMS integration layer, Azure Cosmos DB or Azure SQL for posting log. Programming languages: TypeScript/React for the PWA, C# (.NET 8) or Node.js for the backend. Basic enablement for Azure and GitHub Copilot is recommended.

Candidate profile

  • Bachelor or Master students in Computer Science, Software Engineering or related fields. Useful knowledge: web/mobile development (TypeScript, React or similar), basic understanding of REST APIs and cloud services, interest in conversational AI and UX. No prior Azure experience required – enablement will be provided.

Additional information

SPG will provide a mocked backend of the SPG Warehouse Suite (REST API, sample articles, bin master data) plus realistic goods-receipt test cases. Bluetooth headsets for the demo will be made available by SPG.

Motivation

Smart City AMR systems enable automated healthcare logistics—freeing time for patient care.

Target

Introduction of an AMR fleet management system and extension with AI-based decision-making mechanisms for the optimization of hospital processes.

Approach

  • From code to reality: Develop smart AI components that transform an existing AMR system into an intelligent SmartCareLogistics platform

Actors/Stakeholders

Healthcare staff, patients, operators, technology partners

Business Value

Improved process efficiency, better coordination, higher service quality

DoD/Success Metrics / KPIs

Workload reduction for healthcare staff, Reduction of time per task

Distribution of Activities

  • 50% development & integration
  • 30% testing
  • 20% presentation

Technical prerequisites

Java, Angular, AI/ML basics, system integration

Candidate profile

  • Students in Computer Science, AI, Robotics (Bachelor/Master)

Target

Detection of crop rows using camera/image data in heavily weeded fields

Distribution of Activities

  • 30 % development and programming
  • 10 % user experience design
  • 60% AI training and integration

Technical prerequisites

VS, VSCode, Microsoft Azure Services, GitHub, Copilot

Candidate profile

  • Python, (C++), GIT); Basic AI understanding (e.g. YOLO

Motivation

Physical AI and humanoid robots are progressing rapidly, but industrial companies still lack practical methods to translate human work instructions into safe, structured and executable robot actions. i-LogiX wants to explore how AI-based task planning can reduce this gap and make humanoid applications more tangible for industrial use.

Target

Develop and evaluate a prototype for LLM-supported task planning for humanoid robots in an industrial scenario. The system should translate natural-language work instructions into structured subtasks and map them to executable or simulated robot actions.

Approach

  • Select a simple industrial humanoid robot task and explore the available robot capabilities, tools, data, and infrastructure. Build a prototype that uses an LLM to convert natural-language instructions into structured subtasks and map them to executable or simulated robot actions. Validate whether the generated task plans are complete, technically usable, and executable in the simulation environment.

Actors/Stakeholders

i-LogiX, industrial companies evaluating humanoid robot applications, technical decision-makers

Business Value

The use case reduces uncertainty around the industrial deployment of humanoid robots by demonstrating how AI-based task planning could work in practice. The results can support future decisions and provide reusable knowledge, software components, prompt strategies, or evaluation criteria for later robotics projects.

DoD/Success Metrics / KPIs

The challenge is successful if students deliver a tangible result that shows how natural-language instructions can be transformed into structured robot task plans and technically validated.

Distribution of Activities

  • 50% development and programming
  • 50% AI training and integration

Technical prerequisites

Azure OpenAI + Copilot

Candidate profile

  • The challenge is successful if students deliver a tangible result that shows how natural-language instructions can be transformed into structured robot task plans and technically validated.

Additional information

Due to the complexity of humanoid robots and their simulation under time constraints, conceptual and architecture-focused results are also valid if they clearly support future technical implementation.

Motivation

The DiLab at the University of Passau supports educational innovation in teacher education through microteaching a structured approach where (prospective) teachers practice and reflect on core teaching skills in recorded sessions. Currently, the reflection process depends on disconnected tools and lacks AI-based support. The goal is a single, flexible digital solution that enables both peer and AI-assisted reflection, guided by user-defined observation criteria, for any microteaching scenario.

Target

We aim to develop a digital all-in-one solution for effectively organizing microteaching in teacher education

Approach

  • Designing and testing a web-based, all-in-one prototype that structures the microteaching reflection process and integrates
  • Flexible video integration: Secure upload and playback of recorded microteaching sessions.
  • Custom criteria definition: Users can set their own observation criteria that guide both peer and AI feedback.
  • Time-stamped peer feedback: Peers can leave comments linked to specific video moments, aligned with the defined criteria.
  • Criterion-based AI analysis: Integration of a generative AI model that uses the video transcript and user-defined criteria to generate structured, criterion-based feedback.

Actors/Stakeholders

(pre-service) teachers, university instructors, DiLab team in teacher education

Business Value

one platform replaces disconnected tools; works for in-person and remote sessions alike; consistent, criteria-based reflection at scale; adaptable to any subject

DoD/Success Metrics / KPIs

all 4 core modules integrated in one UI; AI workflow demonstration: successful end-to-end run of a video material and feedback (collaborative, AI) workflow within the Web Application; All external libraries are fully documented, and the project includes clear, accurate references and attributions to all copyrighted libraries and their respective copyright owners (licensing)

Distribution of Activities

  • 45% development and programming
  • 25% AI integration
  • 20% user experience design
  • 10% testing

Technical prerequisites

Programming of web applications using any suitable programming language or framework of your personal choice (e.g Laravel, Next.js); GitLab based workflow / collaborative version control tools e.g (Git); Use of Open Source LLMs for integrating AI-enhanced functions; Comfort with modern package managers and build tools (e.g. npm, pnpm, pip, Turbopack)

Candidate profile

  • Basic WebDevelopment, Software Engineers, Python, Javascript, SQL, UX/UI Design, Background or genuine interest in pedagogical practice and/or educational research

Additional information

(Provide any additional information, if required): The prototype is intended as a functional proof-of-concept, not a production system. Security hardening, user management, and LMS integration are potential next steps after the innovation challenge. The DiLab team is available to provide sample video material and example observation criteria for testing during the innovation challenge.

Motivation

In a modern Smart City, the quality of life depends significantly on functioning infrastructure (transport, energy, public spaces). Currently, faults or disruptions are often reported in a fragmented, incomplete, or purely colloquial manner through various channels (citizen reports, sensors, technician notes). These "data silos" prevent efficient, AI-based coordination. The challenge is to capture and process this unstructured information in real-time so that it becomes usable for automated city processes and AI-based analytics.

Target

The goal is to turn urban infrastructure management into a resilient, data-driven operation. The objective of this challenge is to develop a prototype that:
  • Actively assists in creating incident reports and action logs to ensure the capture of complete and structured information.
  • Stores this information in a machine-readable format (e.g., based on schema.org or urban data standards).
  • Provides role-based user interfaces for citizens, emergency responders, and city planners.

Approach

  • During the 24h challenge, a prototype will be developed to address the following core tasks:
  • Data Quality: Development of an AI assistant that helps users formulate reports (e.g., "broken street light" or "water pipe burst"), ensuring all relevant fields are filled and information is captured structurally as schema.org-compliant objects.
  • Usability: Design and prototypical implementation of UIs tailored to specific actors: an intuitive mobile app for citizens/field workers and an expert dashboard for the central command center.
  • Workflow Modeling: Mapping the entire Smart City process – from event generation and AI-supported prioritization/routing to the relevant departments, through to activity documentation and status tracking.

Actors/Stakeholders

  • Mobile Task Force: Technicians and service providers handling tasks on-site.
  • Central Command Center: Coordinators managing resources and maintaining a city-wide overview.
  • Smart Citizens: Engaged residents who want to report issues easily and receive transparent feedback.

Business Value

  • Improved response times for critical infrastructure failures.
  • High-quality data as a foundation for long-term urban planning (Predictive Maintenance for the city).
  • Sustainable preservation and machine-readable preparation of institutional knowledge.
  • Increased citizen engagement through transparent, digital feedback loops.

DoD/Success Metrics / KPIs

  • Ratio of structured, schema.org-compliant reports vs. unstructured text entries.
  • Reduction in processing time due to better initial data capture.
  • User satisfaction with the new interfaces (Usability feedback).
  • Proof of concept for AI-driven automation (e.g., automated department assignment).

Distribution of Activities

  • 35% Development & Backend (Workflows, Data Model, API).
  • 35% User Experience & Interface Design (Mockups, Prototyping, Usability Tests).
  • 30% AI Integration & Data Modeling (schema.org, Prompt Engineering, Validation).

Technical prerequisites

  • AI Platform: Halerium or similar platforms for AI-driven process automation and knowledge management.
  • Experience in Prompt Engineering and understanding of AI Agent architectures.
  • Use of AI-supported code generation (e.g., Manus, GitHub Copilot, etc.) to accelerate development.
  • Experience in responsive UI development (e.g., React, Flutter, or similar).

Candidate profile

  • Students/Graduates (Computer Science, Business Informatics, Urban Planning with an IT focus).
  • Interest in Cloud Architectures, Smart City technologies, and data standardization.
  • Hands-on mentality for rapid prototyping.

Additional information

The focus is on how intelligent assistance and structured data modeling (schema.org) can create the foundation for powerful, generative AI systems in urban administration and how practical interfaces can increase acceptance and data quality in everyday city life.

Motivation

Stadtwerke Plattling ensure the reliable supply of electricity, water, wastewater, and heating, thereby operating core components of critical infrastructure within a Smart City / Smart Region context. In the event of service disruptions, the primary focus lies on restoring supply as quickly as possible. At the same time, there is a strong need for transparent and timely communication with citizens. In practice, staff in standby and fault resolution services are heavily occupied with the technical handling of incidents. As a result, the preparation of citizen information or press communications is often delayed. This leads to information gaps, inconsistent messaging, and additional workload for operational staff. Citizens, however, expect clear, fast, and easily understandable information regarding the nature, scope, and expected duration of service disruptions. Against this background, the Innovation Challenge calls for the development of a prototypical solution that supports operational staff in the rapid and standardized creation of outage-related communications.

Target

The objective of this use case is the design and implementation of an intelligently supported application for the rapid and standardized creation of outage communications in utility operations. Staff shall be able to report incidents via a mobile device, for example through speech input, structured forms, or predefined selection options. The system shall process these inputs automatically, structure relevant information, and generate an appropriate communication draft. The technical realization of this sup-port is intentionally kept open and may be implemented in a rule-based, form-driven, or automated processing and recommendation-based manner. The goal is explicitly not the use of the most complex technology stack, but rather the development of a fast, robust, and operationally viable solution for real-world outage communication. The generated draft shall be presented to staff for review and approval. Upon approval, the information shall be automatically published through existing communication channels. The entire process, from incident capture to publication, should ideally be completed within two to three minutes.

Approach

  • The technical implementation shall be designed independently by the student teams. A structured processing chain is expected, in which an incident is first captured and subsequently processed in an automated manner. Relevant information such as type of incident, location, affected utility service, impact, and estimated duration shall be identified and structured for further processing. Based on this information, a communication draft shall be generated automatically. Teams are free to choose their technical approach, including structured forms, rule-based systems, template-driven generation, spe-ech processing, automated text generation mechanisms, or hybrid approaches. The final draft shall be displayed in a user interface for review and approval. After approval, the information shall be transmit-ted to existing communication channels. For prototypical integration with the Heimat-Info-App or exis-ting publication systems, several options are available. These include REST API interfaces, RSS feed integration, or embedding via iFrame technologies. In addition, manual publication via an editorial ba-ckend may also be used. The choice of integration approach is part of the student solution design.

Actors/Stakeholders

The primary actors in this process are internal staff of Stadtwerke Plattling, in particular personnel from standby and incident response services who capture and approve outage reports. Additional internal stakeholders include communication and customer service teams, who benefit from faster and more consistent information provision. External stakeholders include citizens who require timely and understandable information during service disruptions, as well as municipal administrations and media representatives who depend on consistent and reliable communication.

Business Value

The proposed solution reduces manual effort in the creation of citizen-facing communications and relieves operational staff during critical situations. At the same time, it increases the speed of informa-tion dissemination and improves the quality and consistency of communication. By introducing stan-dardized processes and automated support mechanisms, transparency towards citizens is enhanced and the number of incoming inquiries is potentially reduced. The use case also addresses key aspects of modern Smart City architectures, including critical infrastructure resilience, digital citizen communi-cation, and intelligent process support.

DoD/Success Metrics / KPIs

The use case is considered successfully implemented if incidents can be efficiently captured, automa-tically processed, and transformed into appropriate communication drafts, with a clear approval work-flow in place. Key evaluation criteria include time from incident capture to publication, clarity and readability of generated communications, usability of the system, and transparency of the technical approach. Additional technical KPIs may include the accuracy of automated classification, correctness of information extraction, and quality of generated communication drafts.

Distribution of Activities

The implementation of the use case is to be carried out independently by student teams. A reasonable division of responsibilities is expected across frontend/application development, backend develop-ment, UX/UI design, data and process modeling, and automated processing logic. The focus lies on delivering a functional and demonstrable prototype that showcases both technical feasibility and user-centered design.

Technical prerequisites

The technical solution shall be designed by the students themselves. The focus is on developing a prototypical but realistic system within a Smart City architecture context. Suitable technologies for data management, automated processing, frontend and backend development, and potential speech or text processing components shall be selected and integrated independently. The system shall in-tegrate with NAME1 communication channels via suitable interfaces or existing publication pathways. For prototypical integration with the NAME2 app or related systems, several options are available, including REST API interfaces, RSS feed distribution, and iFrame-based integration. Alternatively, manual publication via an editorial interface may also be used. The implementation may be fully proto-typical or simulated; full production integration is not required. Data protection, transparency, and practical applicability shall be considered in the system design.

Candidate profile

  • This use case targets students in Computer Science, Business Informatics, Data Science, or related fields. Useful skills include software development, web or application development, APIs, data proces-sing, and a basic understanding of digital process automation and modern system architectures. Fa-miliarity with UX/UI design and cloud-based systems is beneficial but not mandatory.

Additional information

The language of the Innovation Challenge is English. All deliverables, documentation, and presentati-ons must therefore be prepared in English. Teams consist of five students per use case. The objective is to develop an innovative, practice-oriented, and presentation-ready prototype within the given timeframe. No intensive supervision is planned. Students are expected to work independently, in a structured and solution-oriented manner. The use case is intentionally designed to allow for diverse technical approaches and creative implementations, without drifting into full production system com-plexity.

Motivation

A single administrative update currently requires rewriting and reformatting for 5–8 different communication channels. This consumes scarce staff time in small municipalities. An AI assistant can automate this repetitive work and free capacity for strategic communication.​

Target

Automatically transform one short administrative input into ready‑to‑publish texts for all relevant channels. Ensure consistent tone, target‑group alignment, and visual output. ​

Approach

  • Build an AI assistant that turns one short administrative input into ready‑to‑publish multi‑channel content, including text and simple branded visuals.​

Distribution of Activities

  • 35% AI Integration & Prompt Engineering
  • 30% Frontend & UX
  • 20% SharePic Generator
  • 15% Content, Testing & Presentation

Technical prerequisites

  • LLM API (OpenAI, Claude, or similar)
  • Web Stack: HTML, CSS, JavaScript, or React
  • PDF Processing (PDF.js or Python)
  • Graphics Generation (HTML Canvas, html2canvas, or Image API)
  • Optional Azure Services: Azure OpenAI, Azure App Service, Azure Blob Storage​
  • Optional Integration with the Municipal Software & AI Platform Ayunis

Motivation

Economic development agencies support companies, investors, municipalities and regional partners in addressing a wide range of economic challenges. At the same time, the volume of relevant information is continuously increasing, including:​
  • new funding programs​
  • economic developments​​
  • investments and company expansions​​
  • business relocations and settlements​
  • industry trends
  • workforce and skills-related topics​
  • energy and transformation issues​
  • public discussions and emerging regional topics​
Today, this information is scattered across numerous websites, databases, media portals and information services. Research is often conducted manually and requires significant time and effort. As a result, important developments may only be identified at a later stage, while valuable resources for strategic activities are lost. Economic development agencies therefore need new ways to automatically collect relevant information, intelligently analyze and prioritize it, and derive actionable insights and recommendations from it.

Target

Development of an AI-powered Economic Intelligence Radar for the district of Dingolfing-Landau.​
The system should automatically collect, analyze, prioritize and present publicly available information in a clear and understandable way.​
The objective is to generate a digital Economic Journal on a regular basis – ideally every month – summarizing current economic developments, relevant funding opportunities, opportunities and risks, industry trends, and significant regional business events for the economic development agency.​
In addition, the system should derive concrete and prioritized recommendations for action based on the insights generated. Ideally, these recommendations should be based on economic indicators, regional structural data, current developments, and scientifically sound analytical and evaluation methods.​
The goal is to help identify opportunities at an early stage, make risks more visible, and support strategic decision-making through data-driven insights and evidence-based recommendations.​

Approach

  • Development of an intelligent research and analytics system that automatically evaluates various information sources and generates economically relevant insights.
  • Funding Program Monitoring​ - Identification of new funding programs, - Analysis of changes to existing programs, ​ - Assessment of relevance for companies within the district​
  • Economic Intelligence Radar​ - Detection of investments, - Company developments, - Business expansions, ​ - Corporate anniversaries, - Workforce and employment developments, - Public discussions and emerging topics, - Economic and industry trends, - Regional structural and economic data​
  • AI-Powered Economic Journal Automated generation of a report containing:​ - the most important developments from the previous reporting period, - relevant funding opportunities, - regional trends, - opportunities and risks, - prioritized recommendations for action​
  • Based on the insights generated, the AI should derive concrete recommendations for the Economic Development Agency. Examples include: - proactively informing companies about relevant funding, - opportunities recommending direct outreach to companies planning investments or expansions, - suggesting new event formats or support initiatives, identifying and highlighting emerging industry developments, - detecting support needs and economic challenges at an early stage, - identifying strategic opportunities and risks for the regional economy. The ultimate goal is not only to collect and summarize information, but to transform data into actionable intelligence that supports evidence-based decision-making and proactive economic development.​

Actors/Stakeholders

  • Economic Development Agency of the District of Dingolfing-Landau
  • Companies and businesses
  • Investors
  • Start-ups and entrepreneurs
  • Municipalities
  • Chambers and business associations (e.g. Chamber of Industry and Commerce – IHK, Chamber of Skilled Crafts – HWK)
  • Regional network partners​

Business Value

For the Economic Development Agency​
  • Reduction of manual research efforts
  • Better overview of economic developments
  • Early identification of relevant trends and opportunities
  • More informed and evidence-based decision-making
  • Increased ability to act proactively
  • More targeted and audience-specific communication
For Companies​
  • Faster access to relevant funding opportunities
  • Better visibility of regional support services and programs,
  • Earlier awareness of economic opportunities and market developments
  • Targeted communication and knowledge transfer tailored to specific needs and industries​
For the Region​ - , - , - , - , -
  • Increased regional competitiveness
  • Improved networking and integration of economically relevant information
  • Reduction of information overload through intelligent prioritization of relevant content
  • Stronger innovation capacity and regional development
  • Enhanced attractiveness as a business and investment location​

DoD/Success Metrics / KPIs

  • Reduction in research and information-gathering time
  • Number of funding opportunities identified
  • Number of economic developments detected
  • Quality and relevance of generated recommendations for action
  • User satisfaction
  • Number of regularly generated reports
  • Number of measures, initiatives or activities triggered by the generated insights​

Distribution of Activities

  • 35% Artificial Intelligence & Data Analytics
  • 25% Data Research & Data Integration
  • 20% Software Development
  • 10% UX/UI Design & Data Visualization
  • 10% Business Concept & Presentation

Technical prerequisites

  • Artificial Intelligence (AI)
  • Data Science
  • Information Management
  • Web Scraping & Data Integration
  • Software Development
  • Business Informatics
  • Business Intelligence​

Additional information

The challenge is not primarily about developing yet another database, but about the intelligent integration, evaluation and presentation of information that is already available from a wide range of sources. Potential information sources include:​
Funding Programs​ - Federal Funding Portal (Förderzentrale des Bundes), - Bayern Innovativ, - Funding programs provided by the European Union, the Federal Government and the State of Bavaria​
Economic Developments​ - Regional newspapers and local media, - National and international media sources, LinkedIn, - Corporate websites, Press releases, - NiWi – the IHK business magazine, ​ - Chamber of Industry and Commerce (IHK) Lower Bavaria, - Chamber of Skilled Crafts (HWK) Lower Bavaria–Upper Palatinate​
Structural and Economic Data​ - IHK location and business statistics, Statistics Bavaria, Federal Employment Agency (Bundesagentur für Arbeit), Regional economic databases, Open Data portals​
Regional Information​ Municipal websites, District business registry and company database (KWIS, potentially using anonymized sample data), Press portals, Economic and business news sources​
*In addition to publicly available sources, the Economic Development Agency can provide insights into established research processes, commonly used platforms and relevant information channels that have proven valuable in daily economic development practice.​

Motivation

To simulate and train robotic systems—particularly those based on Physical AI, such as humanoid robots—the entire factory environment must be modeled in accordance with real-world physical conditions.​
In industrial settings, detailed CAD data of factory environments is typically available; however, this data often lacks physical attributes. Assigning appropriate physical parameters to these datasets is time-consuming, costly, error-prone, and highly repetitive.​

Target

The goal of this project is to select and adapt an AI-based solution that can automatically identify specific objects within CAD models. In parallel, the AI system should extract relevant product data—especially physical properties—from supplier and manufacturer databases, and accurately match this information to the corresponding CAD objects.​

Approach

  • Python (within NVIDIA Isaac Sim)​
  • Visual Studio Code​
  • On-premise AI models (no proprietary CAD data will be uploaded to external cloud services)​

Distribution of Activities

A pilot workstation is successfully reconstructed in NVIDIA Isaac Sim based on provided CAD data, with material and physical properties automatically assigned to the corresponding CAD objects using AI.​

Motivation

Banks in the DACH region operate under an ever-growing volume of regulatory documentation (BaFin circulars, MaRisk, EBA guidelines, internal policies). Compliance, risk and second-line-of-defence teams spend significant time searching, cross-referencing and quoting these documents. Public LLM services (e.g. ChatGPT) cannot/shouldnot be used because documents and questions are highly confidential and subject to banking secrecy.​

Target

Build a privacy-preserving AI assistant that lets a bank employee ask natural-language questions on a curated body of regulatory and internal documents and receive precise, citation-backed answers, with the LLM and all data running inside infrastructure fully controlled by the bank. A core deliverable is a comparative evaluation of two deployment architectures (managed vs. self-hosted) with a clear recommendation.​

Approach

  • Curate a sample corpus of public regulatory documents (BaFin, MaRisk, EBA) as the demo dataset.​
  • Build an ingestion pipeline (PDF parsing via Azure AI Document Intelligence, chunking, embedding) and index in Azure AI Search (vector + hybrid).​
  • Implement two RAG variants in parallel: (A) Managed – Azure OpenAI Service (GPT-4o) inside the customer’s Azure tenant via Private Endpoint; (B) Self-hosted – open-source LLM (e.g. Llama 3.3 70B or Mixtral 8x22B) deployed on Azure ML Managed Online Endpoints or AKS with GPU nodes in the customer’s tenant. Model weights live on tenant-controlled storage.​
  • Develop a chat-style web UI with source highlighting, confidence indicator and audit trail.​
  • Benchmark both variants on the same evaluation set: answer accuracy, latency, cost per 1k queries, regulatory posture (where do weights run, where does data flow, AVV implications).​
  • Document the decision matrix “Managed vs. Self-hosted” as guidance for regulated industries.​ ​

Actors/Stakeholders

Compliance officers, risk managers, internal audit, IT security and data-protection officers of a typical DACH bank. Internally: SPG Germany product/sales team driving the “Private AI” offering, SPG nearshore development team.​

Business Value

Tangible proof-of-concept and reference architecture for SPG’s “Private/On-Premise AI” go-to-market story in regulated industries, immediately reusable in concrete sales situations with banks and asset-finance institutions. The comparative evaluation provides a defensible answer to the typical customer question “why not just use ChatGPT?”. Productivity gain in compliance teams estimated at 20–30% on document-research tasks.​

DoD/Success Metrics / KPIs

Working web UI deployed on Azure; both RAG variants (managed and self-hosted) operational; ingestion of at least 50 regulatory documents; answers with at least 2 verifiable source citations per query; answer accuracy ≥ 85% on a prepared question set (graded by coaches); side-by-side benchmark table for both variants; architecture documentation with a clear “data & weights stay in tenant” statement for the self-hosted variant.​

Distribution of Activities

  • 40% development and programming (ingestion pipeline, RAG backend, chat UI)
  • 30% AI training and integration (deployment of open-source LLM, prompt engineering, retrieval tuning, benchmark)
  • 15% architecture and security documentation (private-deployment story, AVV/MaRisk angle)
  • 15% user experience design (citation display, audit trail, comparison view).

Technical prerequisites

Azure AI Search (vector + hybrid), Azure Blob Storage, Azure AI Document Intelligence (PDF parsing), Azure App Service or Container Apps (web UI), Azure Key Vault, Azure Virtual Network with Private Endpoints. For variant A: Azure OpenAI Service (GPT-4o, embeddings). For variant B: Azure ML Managed Online Endpoints or Azure Kubernetes Service with GPU nodes (NC-series), running an open-source LLM (Llama 3.3, Mixtral or similar) plus open-source embedding model (BGE-M3 or E5). Programming languages: Python (FastAPI) for the backend, TypeScript/React for the frontend. Basic enablement for Azure, Azure ML and GitHub Copilot is recommended.​

Additional information

SPG will provide the curated sample document corpus, a prepared evaluation question set and access to an Azure subscription with GPU quota. The deliverables (reference architecture and benchmark) will feed into SPG’s “Private AI” reference architecture for regulated industries.​
Candidate profile: Bachelor or Master students in Computer Science, Data Science or Information Systems. Useful knowledge: Python, basic NLP/LLM concepts, web development, interest in regulatory/compliance topics and cloud architecture. Prior exposure to vector search, RAG or model deployment (Hugging Face, vLLM) is a plus but not required – enablement will be provided.​

Motivation

Large public events such as the Volksfest and Weinfest attract thousands of visitors to Landau an der Isar every year. Growing attendance increases challenges related to visitor flows, traffic, parking, safety, and resource planning.​
This use case is intentionally open-ended: the goal is not a finished product, but innovative, practical AI-driven concepts that help cities plan, manage, and optimize large events. The main focus is the analysis, prediction, and optimization of visitor flows.​

Target

Develop a software, web, or app concept focused on visitor flow analysis and optimization.

Approach

  • Possible solutions: Dashboard, heatmaps, digital twin, real-time monitoring, simulation, visitor assistant​.
  • Data sources: Historical visitor data, previous visitor tracking, safety data, traffic/parking data, weather, event data, geodata, sensors, anonymized camera or WLAN/Bluetooth data​.
  • AI components: Forecasting, anomaly detection, pattern recognition, simulation, recommendations.
  • Creativity, practicality, and modern technologies; teams are encouraged to go beyond the listed examples​.

DoD/Success Metrics / KPIs

Better event operations, improved safety, and data-driven planning for future events​

Technical prerequisites

Tech stack: Azure, Python, JavaScript/TypeScript, SQL, REST APIs, Power BI/Grafana, OpenStreetMap/Leaflet/Mapbox

Candidate profile

  • Students or young professionals in software, AI, data science, geodata, web, or UX/UI​

Motivation

Support customers in identifying unusual energy and water consumption early – reducing costs and conserving resources.​

Target

Identify unusual consumption patterns, ​assess possible causes and derive actionable recommendations.​

Approach

  • Compare data with historical, seasonal and regional benchmarks.​
  • Use AI- or rule-based anomaly scoring; provide heatmaps and dashboards.
  • Data Sources: Anonymized electricity, gas and water consumption; ​PV feed-in; weather and solar data; asset and fault reports.​

Business Value

Greater customer value and loyalty, proactive advice, fewer customer​ enquiries, and smarter use of existing data.​
Expected Impact: Lower costs and CO₂ emissions, earlier detection of defects or leaks,​ and greater transparency on regional consumption trends.

Motivation

Make everyday interactions with Stadtwerke Landau simpler, more transparent and more convenient for customers.​

Target

Conceptual design of a customer-centric Stadtwerke app that bundles​ relevant services and local information.​

Approach

  • Identify use cases; define user journeys and a feature roadmap;​ create wireframes or a clickable prototype.​
  • Data Sources: consumption, generation and billing data; customer account data; service notifications and local information.​

Business Value

Improved customer experience and loyalty, fewer service enquiries,​ stronger digital customer channel, and better visibility of Stadtwerke​ services.​
Expected Impact: Faster access to information, higher service quality, greater transparency on consumption and bills, and closer customer relationships.​

Motivation

The company „Unternehmensgruppe Dr. Mirski“ operates two in-house kitchens that prepare fresh meals every day for residents across their eight care faciliites. Planning menus is a complex task as it must consider nutritional guidelines for elderely people, allergies, medical diets, personal references, seasonal ingredients, purchasing costs and the efficient use of available food.​

Target

A developed AI-powered meal planning assistant that supports the company by automatically creating optimized weekly meal plans while considering:​
  • Personal references​
  • Allergies, intolerances, dietary requirements and diets​
  • Reduction of food waste/ Improving sustainability​
  • Operational efficiency/ Support​

Approach

  • This use case explores how AI can support the kitchen team and the staff in the care facilities by generating optimized meal plans that improve efficiency, reduce food waste and ensure high-quality (costs – sustainability – digitalisation – satisfied residents)​

Actors/Stakeholders

Kitchen staff, employees in care facilities, residents, controlling​.

Business Value

Reduce food waste and disposal costs, reduce time, all allergies, dietary restrictions correctly considered and easily changed, budget planning (predefined budget limits), better utilization of available ingredients, cost saving, sustainable initiatives within care facilities​.

DoD/Success Metrics / KPIs

  • Reduced time required for menu planning​
  • Meal plans remain within predefined budget limits​
  • Estimated reduction in food waste compared to current planning process​
  • All resident requirements correctly consideres und adaptable​
  • AI generates a complete weekly meal plan automatically​

Technical prerequisites

Master of Arts or Bachelor of Business Administration.