new file mode 100644
index 0000000..3d165cd
@@ -0,0 +1,24 @@
+---
+type: idea
+title: college essay vector grader
+tags: [education, ai, embeddings, software]
+status: raw
+first_captured: 2026-04-10
+updated: 2026-04-10
+sources:
+ - sources/google-sheets-ideas.md
+---
+
+# college essay vector grader
+
+uses vector embeddings to compare essay drafts against successful essays with live distance-to-target metrics. the transparency of showing "how far am I from good" is the novel UX — unlike LumiSource and GradGPT which grade essays using AI trained on successful essays but don't expose vector embeddings or distance metrics to the user.
+
+needs a corpus of successful essays (legally tricky). the embedding visualization is the differentiator. could use open-source embedding models. connects to the [[ultimate-describer|precision description engine]] for the writing quality improvement angle and to [[oncue|OnCue]] for the application writing use case.
+
+**spreadsheet evaluation:** originality 7/10, excitement 7/10, MVP 2-4 weeks. tech depth 5/10, labeled [WRONG FIT] ("embeddings + cosine similarity — could be done in a weekend, not a month").
+
+---
+
+## timeline
+
+- [2026-04-10] captured from google sheets — 7/10 originality, 7/10 excitement, [WRONG FIT] tier