Kl/scrum 36 chats (#51)

Co-authored-by: kevin-lann <[email protected]>

Author: kevin-lann

course-matrix/backend/package-lock.json

@@ -26,6 +26,7 @@
     "express": "^4.21.2",
     "langchain": "^0.3.19",
     "openai": "^4.85.4",
+    "pdf-parse": "^1.1.1",
     "swagger-jsdoc": "^6.2.8",
     "swagger-ui-express": "^5.0.1",
     "ts-node": "^10.9.2",

course-matrix/backend/package.json

@@ -39,11 +39,12 @@ export const yearToCode = (year: number) => {
 
 // Set minimum results wanted for a similarity search on the associated namespace.
 export const namespaceToMinResults = new Map();
-namespaceToMinResults.set("courses", 10);
+namespaceToMinResults.set("courses_v2", 10);
 namespaceToMinResults.set("offerings", 16); // Typically, more offering info is wanted.
 namespaceToMinResults.set("prerequisites", 5);
 namespaceToMinResults.set("corequisites", 5);
 namespaceToMinResults.set("departments", 5);
+namespaceToMinResults.set("programs", 5);
 
 // Consider the last X messages in history to influence vector DB query
 export const CHATBOT_MEMORY_THRESHOLD = 3;

course-matrix/backend/src/constants/constants.ts

@@ -1,6 +1,6 @@
 // Keywords related to each namespace
 export const NAMESPACE_KEYWORDS = {
-  courses: [
+  courses_v2: [
     "course",
     "class",
     "description",
@@ -57,16 +57,8 @@ export const NAMESPACE_KEYWORDS = {
     "concurrent",
     "co-enroll",
   ],
-  departments: [
-    "department",
-    "faculty",
-    "program",
-    "major",
-    "minor",
-    "specialist",
-    "division",
-    "academic unit",
-  ],
+  departments: ["department", "faculty", "division"],
+  programs: ["program", "major", "minor", "specialist", "degree", "stream"],
 };
 
 // General academic terms that might indicate a search is needed

course-matrix/backend/src/constants/promptKeywords.ts

@@ -58,8 +58,8 @@ function analyzeQuery(query: string): {
 
   // If a course code is detected, add tehse namespaces
   if (containsCourseCode) {
-    if (!relevantNamespaces.includes("courses"))
-      relevantNamespaces.push("courses");
+    if (!relevantNamespaces.includes("courses_v2"))
+      relevantNamespaces.push("courses_v2");
     if (!relevantNamespaces.includes("offerings"))
       relevantNamespaces.push("offerings");
     if (!relevantNamespaces.includes("prerequisites"))
@@ -70,8 +70,8 @@ function analyzeQuery(query: string): {
   if (DEPARTMENT_CODES.some((code) => lowerQuery.includes(code))) {
     if (!relevantNamespaces.includes("departments"))
       relevantNamespaces.push("departments");
-    if (!relevantNamespaces.includes("courses"))
-      relevantNamespaces.push("courses");
+    if (!relevantNamespaces.includes("courses_v2"))
+      relevantNamespaces.push("courses_v2");
   }
 
   // If search is required at all
@@ -83,11 +83,12 @@ function analyzeQuery(query: string): {
   // If no specific namespaces identified & search required, then search all
   if (requiresSearch && relevantNamespaces.length === 0) {
     relevantNamespaces.push(
-      "courses",
+      "courses_v2",
       "offerings",
       "prerequisites",
       "corequisites",
       "departments",
+      "programs",
     );
   }
 
@@ -153,6 +154,8 @@ async function reformulateQuery(
       apiKey: process.env.OPENAI_API_KEY,
     });
 
+    // console.log("History: ", conversationHistory);
+
     // Create messages array with the correct type structure
     const messages: OpenAI.Chat.ChatCompletionMessageParam[] = [
       {
@@ -184,7 +187,13 @@ async function reformulateQuery(
           Output: "What are the course names of course codes: MGTA01, CSCA08, MATA31, MATA35?"
 
           User: "How are you doing today?"
-          Output: "How are you doing today?"`,
+          Output: "How are you doing today?"
+
+          User: "Give 2nd year math courses."
+          Output: "What are some 2nd year math courses?"
+
+          User: "Give first year math courses."
+          Output: "What are some 1st year math courses?"`,
       },
     ];
 
@@ -206,7 +215,7 @@ async function reformulateQuery(
       model: "gpt-4o-mini",
       messages: messages,
       temperature: 0.1, // Lower temperature for more consistent, focused queries
-      max_tokens: 150, // Limit response length
+      max_tokens: latestQuery.length * 3, // Limit response length. Proportional to user input.
       top_p: 0.5, // Reduced top_p for more focused outputs
     });