General AI confidently invents facts. Ask it about your specific course, policies, or exam requirements and it will guess, sometimes dangerously wrong.
It knows a little about everything, but nothing deep about bilingual education, BTLPT competencies, or your institution's specific content.
Without safety layers, it can go off-topic, give inappropriate responses, or be manipulated into ignoring its instructions.
It answers questions. It does not scaffold learning, assess performance, or adapt to the learner's level, the things a good mentor does.
Give the AI access to your own curated knowledge base. It retrieves relevant, verified information before generating a response. The AI still uses its language ability, but the facts come from your data, not its imagination.
What makes RAG truly powerful is how it finds the right knowledge. Instead of keyword matching, the system uses embeddings, a technique that converts text into numerical representations of meaning, so the AI can find relevant content even when the learner's question uses completely different words than the knowledge base.
[0.023, -0.851, 0.442,
0.117, -0.394, ... ][0.023, -0.851, ...][0.019, -0.847, ...]Catches obvious abuse patterns, prompt injection attempts, and inappropriate content before the request even reaches the AI model. Fast, lightweight, first line of defense.
OpenAI's moderation endpoint screens every message for harmful, violent, sexual, or policy-violating content. A second independent filter that catches what regex misses.
Two layers work together: the AI's system prompt enforces behavioral rules (stay on topic, never invent information, say "I don't know" when uncertain, scaffold instead of giving direct answers). On top of that, custom code restricts what topics the conversation can cover, limiting the AI to the specific educational domain it was designed for, regardless of what the user asks.
The vector search is filtered by where the user is on the platform. A homepage visitor cannot access textbook content. A student in a lesson only gets pedagogically relevant material. This protects both accuracy and intellectual property.
Some knowledge is shared across related courses (e.g., bilingual ed. theory applies to both bilingual and ESL contexts), while course-specific content stays isolated.
IP Protection: Expert-created content (book chapters, research papers, articles, and proprietary materials from our specialists) is only retrievable inside the course environment. A casual visitor cannot access it, even if they ask directly.
Custom AI tutors for teacher preparation, certification practice, or academic language development, built on your content and aligned to your standards.
AI-powered professional development tools, scenario-based training systems, and knowledge-grounded chatbots designed for your specific domain.
If you are exploring AI in education and want to collaborate on research, pilot programs, or co-authored work, we are always open to meaningful partnerships.