01 / EXPLORE THE KNOWLEDGE SPACE

Every concept, connected.

Explore relationships between ideas, select a concept, and move from the map into a source-grounded learning session.

LEARNING UNIVERSE
Awaiting course material

Selected concept

Concept

Explore this concept in your course material.

Indexed conceptRelated conceptUnexplored

DRAG TO ORBIT · SELECT A NODE TO INSPECT

Indexed passages

—

Source locations preserved

Concepts found

—

Heuristic concept extraction

Practice attempts

—

Recorded in this workspace

Observed accuracy

—

Not a learning-effectiveness score

Concept index

Select a concept to focus the universe and open its learning path.

0 concepts

Your universe is waiting.Add a PDF, slide deck or notes to extract concepts and build the graph from your own material.

Begin with evidence

Build a knowledge base from material you trust.

Drop files or browsePDF · PPTX · DOCX · TXT · MD · VTT · SRT

Indexed source files will appear here.

02 / SOURCE-GROUNDED TUTORING

Understand the why.

Ask questions about indexed material. Inspect source locations and challenge an explanation when the evidence is insufficient.

Learning conversation

Answers are retrieved from the current course material, not generated from an unverified global knowledge base.

Evidence-first

LUMEN / TUTORAdd course material, then ask a question. I will show the indexed passages used to support an answer and say when evidence is insufficient.

The evidence standard

A fluent answer is not necessarily a supported answer. Inspect the quoted excerpt and its source locator before relying on it.

Grounding policy
If indexed passages cannot substantiate the question, LUMEN should state the limitation instead of presenting an unsupported claim as course fact.

Learning decision trace

After practice, inspect the answer, grading criteria, source location, chosen intervention and any update to the learner estimate.

Current course

No indexed material yet.

03 / RETRIEVAL PRACTICE

Test what you know.

Answer source-derived questions, inspect the grading criteria and review the source location. LUMEN updates a transparent topic estimate after submission.

No active practice session.Index course material, then generate a short practice set. Questions and answer keys are held on the server until submission.

03 / KNOWLEDGE INGESTION

Your source library.

Import supported materials, check ingestion results, and confirm the engine indexed usable text before beginning a grounded session.

Add learning material

Up to 10 files per request; 25 MB per file and 50 MB total.

Source ingestion

Choose course materialsPDF, PPTX, DOCX, TXT, MD, VTT and SRT

Indexed materials

Source filenames and extracted chunk counts returned by the engine.

0 sources

No source material indexed.

Source integrity
Each citation should identify the originating file and page, slide, paragraph or timestamp. Extraction quality still affects answer quality.

04 / INTERACTIVE MATHEMATICS

Think in functions.

Plot a function, change its parameters and inspect how the curve behaves. This self-contained graphing workspace works without an external graphing API key.

Function of x

Supported: x, numbers, + − × ÷, parentheses and powers. Example: 0.5*x^2-2

Horizontal range

−10 to +10

Inspect coordinate

f(2) = 4

This plotter is an educational visualization, not a computer algebra system. It does not perform symbolic integration or guarantee numerical stability for all expressions.

Explore the relationship

Change the expression and coordinate, then predict how the curve changes. Use a course-grounded question in Learn to connect this exploration to your material.

Suggested experiments

05 / OBSERVED LEARNING EVIDENCE

Progress, with context.

Review recorded practice and transparent topic estimates. A prior estimate is not proof of mastery; interpret each score alongside attempts and uncertainty.

Indexed passages

0

Current workspace

Topics detected

0

Heuristic topic labels

Attempts recorded

0

Submitted responses

Observed accuracy

—

Correct / attempted

Topic-level estimate

The current engine uses a simple Beta(2,2)-style prior. It is not a validated Bayesian Knowledge Tracing or Item Response Theory model.

Evidence-linked

Index course material and answer practice questions to populate topic estimates.

Latest learning decision trace

Review the most recent grading result and why the next action is suggested.

A decision trace appears after an answer is submitted.

Interpretation guardrails

1

Estimate ≠ certainty

Confidence intervals are approximate summaries under a simple model.

2

One error ≠ misconception

A diagnosis requires a supported pattern, not a single wrong answer.

3

Accuracy ≠ learning gain

Observed correctness does not establish durable retention or transfer.

06 / SYSTEM VERIFICATION

Evidence of quality.

Inspect live workspace counters and the documented evaluation protocol. Reproducible benchmark results must be generated by the test runner, not fabricated by the interface.

Indexed passages

0Live counter returned by the backend.

Distinct topics

0Heuristic topic labels, not a taxonomy benchmark.

Practice attempts

0Recorded in the current workspace.

Evaluation dimensions

01

Retrieval quality

Measure whether known-answer queries retrieve the expected source chunk; report Recall@k and MRR with the test-set size.

02

Grounding and abstention

Test answerable questions, unrelated questions, misleading term overlap and unsupported answer types.

03

Assessment validity

Test answer keys, numerical tolerances, units and short-answer grading agreement.

04

Adaptation and regression

Check that recorded answers update the expected topic and that changes do not break existing behavior.

Run the test suite

From the project directory, install requirements and run the automated tests. Evaluation results should be regenerated against the exact release candidate.

python -m pytest -q
python evaluate.py

Do not overclaim.
Passing curated fixtures shows behavior on those fixtures only. It does not prove general accuracy or educational effectiveness.