Modal Mind

Modal Mind is an AI-powered personalized learning companion that turns course materials into source-grounded tutoring, adaptive assessments, and mastery-based recommendations.

Comment

Modal Mind is an AI study companion for students who learn from lecture videos, textbooks and slides scattered across different places. It unifies them into one knowledge base where every piece of content stays linked to its origin: a page, a slide number or a video timestamp. The tutor and the quizzes then use that knowledge base and a model of what each student already understands.

The problem Course material is split across videos, PDFs and slide decks, so students waste time searching. General AI chatbots give answers that are not tied to the student's own course material. They also do not know what the student has already mastered, so their help is not personalised. Our solution Multimodal knowledge base. Upload videos, textbooks and slides with no manual preprocessing. The system transcribes video, parses documents, and describes figures and diagrams with a vision model. It organises everything into topics, concepts and prerequisites, and tags each unit with its source location. Source-grounded tutor. Answers use cited excerpts, and each citation opens the exact page, slide or timestamp. If the material does not cover a question, the tutor says so instead of guessing. Any outside knowledge is clearly labelled as outside. Adaptive assessment. Students choose a scope and get quizzes or mock exams with MCQ, short-answer and numerical questions. Every question is tagged with its topic, source location and difficulty. Answer keys are verified by code or by a second model, and repeated questions are blocked. Each answer gets cited feedback, followed by a report of weak topics and likely misconceptions. Learner model. The system keeps a per-topic mastery estimate using Bayesian Knowledge Tracing, updated after every quiz and conversation. A forgetting-curve model sets review timing. New students start with an intake chat and a diagnostic quiz. Evaluation. We measure the system instead of just claiming it works: Retrieval and answer quality are scored with RAGAS or DeepEval: faithfulness, answer relevancy, and context precision and recall. A team-built test set includes off-material queries, to check that the tutor refuses them. Simulated student profiles run across several sessions, and we report mastery gain and question repetition rate. Optional enhancements

Course flow map of topics and prerequisites, targeted revision flashcards, a study schedule built from weak topics and the exam date, and Hindi explanations for English lectures.

Tech stack (proposed)

Python and Streamlit for the app. Speech-to-text, PDF and slide parsing, and a vision model handle ingestion. Hybrid retrieval combines keyword and semantic search. An LLM (Gemini or Claude) writes the grounded answers and questions. The learner model is BKT plus spaced repetition, evaluated with RAGAS or DeepEval.

What makes it different

Tools like Google's NotebookLM already give cited answers and source-based quizzes. Modal Mind adds a persistent learner model, a prerequisite-aware topic map, verified non-repeating questions, and measured evaluation.

Built With

  • ai/ml
  • and
  • i-build-with-python
  • learning
  • llms
  • multimodal-ai
  • pdf/ppt-processing
  • problem
  • rag
  • replit
  • skill
  • solving
Share this project:

Updates

Submission history