From 1d8990590e52067d7ae0103cef2ca040c148dc2e Mon Sep 17 00:00:00 2001
From: ionfwsrijan <ionfwsrijan@users.noreply.github.com>
Date: Fri, 7 Aug 2026 01:03:47 +0530
Subject: [PATCH] fix: route prefix and non-word-character prep topics to the
 AI mentor

The keyword regex used word boundaries on both ends, so prefix keywords
(scal, load balanc, negotiat, probab, math, quant) and keywords containing
non-word characters (c++, c#) never matched. 'Explain probability', 'What is
load balancing?', 'Explain C++ pointers', etc. were misclassified as
off-topic and refused. Anchor keywords as prefixes by dropping the trailing
word boundary.

Closes #1440
---
 ...mainClassifier.prefixKeywords.unit.test.js | 39 +++++++++++++++++++
 backend/utils/domainClassifier.js             |  8 +++-
 2 files changed, 46 insertions(+), 1 deletion(-)
 create mode 100644 backend/tests/domainClassifier.prefixKeywords.unit.test.js

diff --git a/backend/tests/domainClassifier.prefixKeywords.unit.test.js b/backend/tests/domainClassifier.prefixKeywords.unit.test.js
new file mode 100644
index 00000000..bddcc6ef
--- /dev/null
+++ b/backend/tests/domainClassifier.prefixKeywords.unit.test.js
@@ -0,0 +1,39 @@
+import { describe, it, expect } from "vitest";
+import { isPrepPilotDomain } from "../utils/domainClassifier.js";
+
+// ---------------------------------------------------------------------------
+// Domain classifier prefix-keyword fix (issue #1440): core prep topics that
+// are prefixes of real words (probab -> probability, scal -> scaling, math ->
+// mathematics, load balanc -> load balancing, negotiat -> negotiation) and
+// keywords containing non-word characters (c++, c#) previously never matched
+// because of the trailing \b in the keyword regex, so the AI mentor refused
+// legitimate questions about them.
+// ---------------------------------------------------------------------------
+
+describe("isPrepPilotDomain — prefix and non-word-character keywords", () => {
+  const cases = [
+    "probability",
+    "Explain probability.",
+    "What is load balancing?",
+    "How do I scale my database?",
+    "scaling",
+    "Explain C++ pointers.",
+    "Explain C# async/await.",
+    "C++",
+    "C#",
+    "negotiation",
+    "mathematics",
+    "quantitative reasoning"
+  ];
+
+  cases.forEach((query) => {
+    it(`routes "${query}" to the AI mentor`, () => {
+      expect(isPrepPilotDomain(query)).toBe(true);
+    });
+  });
+
+  it("still rejects clearly off-topic prompts", () => {
+    expect(isPrepPilotDomain("Who won the FIFA World Cup?")).toBe(false);
+    expect(isPrepPilotDomain("Write a recipe for pasta.")).toBe(false);
+  });
+});
diff --git a/backend/utils/domainClassifier.js b/backend/utils/domainClassifier.js
index f132d514..976ed79e 100644
--- a/backend/utils/domainClassifier.js
+++ b/backend/utils/domainClassifier.js
@@ -36,7 +36,13 @@ const escapeRegExp = (string) => {
 };
 
 const escapedKeywords = domainKeywords.map(escapeRegExp);
-const keywordRegex = new RegExp(`\\b(${escapedKeywords.join('|')})\\b`, 'i');
+// Anchor keywords as prefixes: keep the leading word boundary but drop the
+// trailing one. A trailing \b made prefix keywords (scal, load balanc,
+// negotiat, probab, math, quant) and keywords ending in non-word characters
+// (c++, c#) never match — "probability", "load balancing", "C++?" etc. were
+// misclassified as off-topic. Prefix matching over-routes a few extra words
+// (e.g. "google" matching "go"), which is acceptable for an AI mentor router.
+const keywordRegex = new RegExp(`\\b(${escapedKeywords.join('|')})`, 'i');
 
 const conversationalRegex = /^(hi|hello|hey|yo|ok|thanks|thank you|who are you|what should i call you|good morning|good evening|\?)$/i;