Language for teaching. Computation for truth.
Every question runs two ways: an ungrounded AI answer you can inspect, and a Wolfram-computed answer that is the source of truth. StepWise compares them and teaches the verified step.
POST /api/askTyped text, or a photo transcription the student reviews first.
Starts both answer paths and waits for the computed result.
Gemini answers from memory with no tools. Fluent, confident — and sometimes subtly wrong. It stays on screen so the exact risk is visible, never hidden.
- 01Intent router
Sends common math straight to a tool; Gemini only translates unfamiliar phrasing.
- 02Wolfram computes
Produces the real result — exact values, numeric values, and the graph.
- 03Gemini narrates
Explains the answer using only the values Wolfram returned.
- 04Number guard
Rejects any number not traceable to the computation, then shows provenance.
Wolfram checks whether the two answers are equivalent. If they differ, StepWise shows the gap (“caught a mistake”); if they agree, it shows an independent confirmation. Either way, the student learns from the computed result.
Results, exact values, graphs, and symbolic equivalence.
Natural-language teaching grounded in returned results.
Unsupported numeric claims fall back to a deterministic computed answer.