Show HN: Educational AI that shows source reliability scores for every response

I built CERAH AI to address the trust problem in AI-generated educational content. Unlike standard AI tools that give you answers without context, CERAH shows exactly which sources inform each response and calculates reliability scores based on source quality.

The system integrates Wikipedia for broad coverage and arXiv for STEM topics, then uses semantic similarity matching to find relevant content. Each response displays a reliability percentage calculated from source types (academic papers weighted higher than blogs) and content relevance scores.

For example, if you ask about quantum mechanics, you'll see whether the answer comes from peer-reviewed papers (high reliability) or general web content (lower reliability), with expandable source details showing similarity scores and direct links.

Current limitations: Uses a small curated knowledge base for core topics, keyword-based related topic suggestions, and mock source references in some reliability calculations. This is an MVP focused on validating whether source transparency actually changes how people evaluate AI-generated educational content.

The question I'm trying to answer: Does knowing that your AI answer is based on Wikipedia vs academic papers vs general knowledge actually influence how much you trust the information? Built with Python/Streamlit, integrates Wikipedia API and arXiv API, uses sentence-transformers for semantic search.

Live demo: https://cerahailearningassistantmvp-bj8fmubn3p3eyu4cohthto.s...

Looking for feedback on whether the reliability scoring concept resonates with the HN community and if the source transparency approach has merit for educational AI tools.