Study smarter with an AI tutor that cites its sources
Upload your lectures, textbooks and slides. Get cited answers when your material has the answer — or a clearly separate general-knowledge response when it doesn’t. Then practice with quizzes that adapt to what you know.
Evaluation not run yet.
Built around the six track requirements
Each capability below maps directly to a numbered requirement in the Track D problem statement.
Multimodal Knowledge Base
Ingest PDFs, slide decks, lecture video, audio and images with no manual preprocessing. Every unit keeps its exact origin.
Source Grounding
Answers grounded in your material cite the page, slide or timestamp. If the material does not cover it, a clearly labelled general-knowledge answer stays separate.
Adaptive Assessment
MCQ, short answer and numerical questions on the scope you choose, verified against your material and tagged to a topic.
Learner Model
Bayesian knowledge tracing plus a forgetting curve gives you a mastery estimate per topic that updates after every answer.
System Evaluation
Faithfulness, relevancy, context precision and recall measured against a test set with known source locations.
Course Map
A visual graph of topics and their prerequisites, colour-coded by your current mastery.
Revision Material
Flashcards and summaries point you back to the topics that could use another look.
Study Schedule
A plan built from weak topics, forgetting-curve decay and how many days are left before your exam.
Hindi / English
Ask in English or Hindi. Hindi explanations keep technical terms in English so nothing gets lost in translation.
Audio Tutoring
Dictate a question with your microphone and have answers read aloud.
Four steps from raw material to mastery
No manual tagging, no spreadsheet of topics. The pipeline builds the structure as it ingests.
- 01
Upload your material
Drop in lecture PDFs, slide decks, recorded lectures or images. StudyMind extracts text, figures and speech with timestamps.
- 02
Knowledge base is built
Content is chunked, embedded, and mapped to topics, subtopics and key concepts — each unit tagged back to its origin.
- 03
Chat or take a quiz
Ask anything and get a cited answer, or generate an adaptive assessment focused on what you have not mastered yet.
- 04
See what changed
Your mastery updates after every answer. The dashboard shows what improved, what is fading, and what to do next.
A retrieval pipeline with a single AI seam
Every AI call goes through one provider module, so the model can be swapped without touching business logic.
Browser → API
JSON over HTTPS with a bearer JWT on every protected route. CORS is locked to the deployed frontend origin.
Ingestion workers
Parse, chunk, embed and topic-tag each source. Unit-level failures are recorded without failing the whole source.
Provider seam
One interface supports Gemini-backed answers and an offline fallback for source-based study.
Trust is a mechanism, not a promise
Cited answers stay grounded, and general answers stay clearly separate.
Cited answers stay grounded; general answers stay separate
The tutor checks your uploads first. When they do not cover a question, any general-knowledge response is shown separately with a clear warning and no citations.
Citations are audited after the fact
Any citation pointing at a unit that was not actually retrieved is stripped before the answer reaches you, and the count of stripped citations is recorded in the evaluation metrics.
Why does the Clock algorithm exist?
The Clock or Second Chance algorithm approximates LRU by keeping a reference bit on each frame. On eviction the hand clears the reference bit and moves on; only a frame whose bit was already clear is actually evicted.
General knowledge — not from your uploaded material: Questions your uploads do not cover can still get a separate general answer.
Illustrative mastery curve
Illustrative values — your real curve is generated from your own answers.
It knows what you know — and what you are forgetting
A per-topic mastery probability, updated after every answer, decayed over time by a forgetting curve.
Bayesian knowledge tracing. Each topic is tracked as a probability of being mastered. A correct answer pushes it up, a wrong one pulls it down — but a wrong answer after long success is treated as a slip, not as forgetting everything.
Forgetting curve. Retention decays exponentially with time. Topics you have not touched resurface on their own in the study plan.
Zone of proximal development. Quizzes deliberately target topics sitting between 30% and 70% mastery — where practice buys the most learning.
Measured, not asserted
Scored against a test set where every question has known source locations, plus deliberately off-material queries.
RAG pipeline metrics
Evaluation not run yet
| Metric | What it measures | Score |
|---|---|---|
| Faithfulness | Answer claims supported by cited excerpts | Evaluation not run yet |
| Answer relevancy | Does the answer address the question asked | Evaluation not run yet |
| Context precision | Retrieved units that are actually relevant | Evaluation not run yet |
| Context recall | Ground-truth units that were retrieved | Evaluation not run yet |
| Citation accuracy | Citations pointing at the correct unit | Evaluation not run yet |
| Outside-material detection | Questions correctly identified as outside your uploaded sources | Evaluation not run yet |
Tech stack
Questions students actually ask
PDF, PPTX, PNG and JPG, MP4 video, MP3 and WAV audio, plus plain text. PDFs are read page by page, slide decks keep slide numbers and speaker notes, and video is transcribed into timestamped segments. Every extracted unit remembers where it came from.
Start studying smarter
Upload your first lecture and see a cited answer in under a minute.