AI Assistant

Hello, how can I help you?

💬
Ask a question...
➤
1
RULE 1
💬 DENMARK & POLAND
2
RULE 2
➔
3
RULE 3
Agenda Setting
Formu- lation
Adoption
Implemen- tation
Evalu- ation
Support & Maint.
🏛️ Policy Cycle
🤖

Welcome to the STRIDE AI Assistant

Get instant, evidence-based answers directly from our European database. To get the most accurate results, follow these 3 golden rules when asking your questions:

Rule 1
Mention Countries by Name
If you want a comparison or specific insights, always write the country names directly in your query. This forces the database to retrieve the exact country chunks immediately.
💡 Try asking: "Compare early language screening for both Denmark and Poland"
Rule 2
Ask for Tables or Bullet Points
Our AI Assistant thrives on structure. Adding formatting preferences prevents long walls of text and gives you clean, professional summaries ready to copy into reports.
💡 Try asking: "Generate a comparison table of barriers and enablers in Norway"
Rule 3
Use Policy Cycle Keywords
The database is structured chronologically using the Policy Cycle framework. You will get precise hits by using official keywords from policy development stages.
💡 Keywords: Agenda Setting, Formulation, Adoption, Implementation, Evaluation

Prompt Examples

Inspiration

Get inspired by these ready-to-use prompts. Copy and adapt them to query the database effectively:

Comparative Overview
“Generate a concise comparative table showing the ‘Agenda Setting’ step for all five national cases: Denmark, Norway, England, Poland, and Hungary. State the core triggers and primary objectives for each.”
Political Recommendations
“What are the key policy recommendations for the ‘Implementation’ stage at the national level regarding the ECE workforce crisis?”
Searching for Specific Mechanisms
“According to the Norway and England case studies, how does physical proximity and assertive outreach (like door-knocking) impact kindergarten enrollment for disadvantaged groups?”
Methodological Insight
“Compare the quantitative research designs and datasets used in the national case studies for Denmark (ITS) and Norway (DiD).”
📋

User Tip: Want to copy generated tables?

You can mark any table directly in the AI Assistant's answer and copy it. Open Microsoft Excel or Google Sheets and paste it directly. You may need to do minor adjustments to columns and rows.

🛠️ Troubleshooting & Quick Fixes
Did the AI Assistant suddenly stop mid-response?
If you request a large comparison or a long list, the chatbot might hit its maximum text length limit for a single response. Don't worry!
👉 Quick Fix: Simply type "Continue" in the chat field, and it will pick up right where it left off.
Is it showing strange formatting or "working notes"?
Because the database compiles many sources, the AI can occasionally output meta-notes (such as Final Polish or Review).
👉 Quick Fix: Type: "Rewrite the response, but omit your working notes and internal reflections."

AI Assistant Chat Window

Overview of the technical setup of the AI Assistant

Below we describe how the Toolbox’ AI Assistant has been built

The toolbox AI Assistant is a RAG-model (Retrieval-Augmented Generation model) and its architecture can be explained this way:

1. Evidence-based data foundation:

Expanded and validated research material comprising peer-reviewed reports (e.g., STRIDE D2.1 and D3.1), legislation, and national case studies.

2. Semi-structured JSON encapsulation (Structured Metadata Indexing):

Each section of the research is tagged with structural metadata (ISCED levels, country names, reform types, years, source references).

3. Vector-based semantic search (Vector Search/RAG):

The AI assistant does not search the internet at random; instead, it derives its answers exclusively from this closed, quality-assured knowledge repository.

Summary
We have a closed, evidence-based RAG repository where high-quality research from PDFs and Word documents has been structured and enriched with metadata (JSON), enabling the vector database to retrieve relevant expert knowledge with pinpoint precision.

The illustration below shows an overview of the AI Assistant is built. It is important to highlight that the AI Assistant only has access to content available and used in the toolbox, including reports, European reform data (also to be found in the Interactive Map), the national cases and briefs.

Illustration of the RAG model used in the toolbox

Unstructured data cover content consisting of ordinary human language text (sentences, paragraphs, reports, analyses, articles). Even though the text in a research report is incredibly well-structured to the human eye (with chapters and headings), computer scientists call it “unstructured” because a traditional SQL database cannot look up an arbitrary sentence as if it were a postal code.

Semi-structured data are a hybrid. We have taken the “unstructured” free text from the reports and wrapped it in a structured container (JSON) with fixed fields: id, chapter, ISCED-level, topics, and text.