Glyph

Background

One-size-fits-all study tools don't adapt to the learner or the subject.

Concept

A learner enters a topic (or uploads a PDF), and Glyph generates a concept map, a curriculum, and an interactive learning "expedition" — cinema-style visual explainers, tutor annotations, and, for STEM topics, derivation walks, sandboxes, simulations, and challenge arenas. A domain classifier gates features so a narrative topic and a calculus topic each get the right components, and difficulty is set from the learner profile.

Build story

The source implementation treats pasted study text as untrusted: concept generation calls a shared sanitizer that strips selected invisible characters and redacts known instruction-override patterns. URL imports reject listed local/private hosts and refuse redirects; PDF extraction checks the file signature and applies a 10 MB upload limit. On sign-in, a claim path attempts to attach anonymous study records to the account, with each table updated separately. Co-built with Audrey; Ben led the agent-driven audit-and-fix waves across the app.

Outcomes

The public home offers the generate-learn workflow, while pricing clearly says online checkout is not switched on. The source includes regression tests that reject unsupported publisher claims, including spaced repetition and concept-level teacher analytics. Account carryover is implemented, but partial failures are possible; this source review does not establish a lossless migration or successful hosted generation.