DESIGN - HUB
Design-Hub Features
Local knowledge base
Local knowledge base (Markdown + embeddings) per subject, selectable on the fly; automatic index rebuild when new files are added.
Semantic search
Semantic search with boosting rules, optional CrossEncoder re-ranking, and neighborhood expansion for coherent context.
Structured answer generation
Structured answer generation (overview/details/practical takeaway), follow-up buttons, feedback logging, and monitoring of user queries.
Process timeline
Process timeline (Discover–Distribute) with keyword highlighting, media cues incl. PDF/image assets, and a web-search fallback when no local matches are found.
Subject and asset management
Subject and asset management, a polished Gradio UI with branding, and download links from assets/.
GDPR Compliance
Built to meet GDPR requirements with privacy-by-design. Data stays in approved regions, is encrypted in transit and at rest, and access is strictly role-based.
Full Data Control
You decide where data lives, how long it’s stored, and who can access it. Bring-your-own storage and keys, on-prem or EU-only hosting, fine-grained RBAC/SSO, and comprehensive audit trails
Quiz-Based Knowledge Checks
Learners can assess their knowledge through built-in quizzes. Difficulty levels are selectable to match the learner’s level, and results are evaluated instantly right after completion.
Our beta version is coming soon
We are currently working hard on our “Design Hub” for mediencollege Berlin. Our focus is on the “Branded Interaction Design” module, which will be expanded step by step in terms of scope and functionality.
Advantages over generic LLMs
Faithful to subject and materials
Faithful to subject and materials: answers are based on curated course content rather than generic model knowledge → higher content reliability.
Transparent sources
Transparent sources: cited passages and assets are shown; users can open downloads/references directly.
Adaptive presentation
Adaptive presentation: timeline highlighting, visual aids, and structured answer formats make it easier for learners to understand.
Context-aware
Context-aware: keyword boosters, re-rankers, and neighborhood logic reduce hallucinations and keep answers close to the original materials.
Feedback loops
Feedback loops: button-based feedback and follow-ups enable fast iterations for instructors—something generic LLMs don’t provide out of the box.
Hallucination Safeguards
With score-based ranking plus booster logic, only evidence-backed matches make it into the prompt. If no suitable context is available, a web fallback kicks in—so the model isn’t tempted to answer from “gut feeling.”