fix: centralize AI model constants and fix invalid model names

src/app/api/AIAssistant/route.ts

@@ -21,6 +21,7 @@ import { executeWebSearchAgent } from "./services/webSearchAgent";
 import normalizeModelContent from "./normalizeModelContent";
 import { withRateLimit } from "~/lib/rate-limit-middleware";
 import { RateLimitPresets } from "~/lib/rate-limiter";
+import { DEFAULT_MODELS } from "~/lib/models";
 
 
 export const runtime = 'nodejs';
@@ -234,7 +235,7 @@ export async function POST(request: Request) {
         }
 
         const embeddings = new OpenAIEmbeddings({
-            model: "text-embedding-ada-002",
+            model: DEFAULT_MODELS.EMBEDDING,
             openAIApiKey: process.env.OPENAI_API_KEY,
         });
 
@@ -376,7 +377,7 @@ export async function POST(request: Request) {
 
         const chat = new ChatOpenAI({
             openAIApiKey: process.env.OPENAI_API_KEY,
-            modelName: "gpt-5.2",
+            modelName: DEFAULT_MODELS.ANALYSIS,
             temperature: 0.7, // Increased for more natural, conversational responses
             timeout: 600000
         });

src/app/api/AIAssistant/services/webSearchAgent.ts

@@ -2,6 +2,7 @@ import { ChatOpenAI } from "@langchain/openai";
 import { SystemMessage, HumanMessage } from "@langchain/core/messages";
 import { performTavilySearch, type WebSearchResult } from "./tavilySearch";
 import { env } from "~/env";
+import { DEFAULT_MODELS } from "~/lib/models";
 
 /**
  * Web Search Agent using LangChain and LangSmith
@@ -91,7 +92,7 @@ async function refineSearchQuery(
 ): Promise<{ refinedQuery: string; reasoning: string }> {
     const chat = new ChatOpenAI({
         openAIApiKey: env.server.OPENAI_API_KEY,
-        modelName: "gpt-5-mini",
+        modelName: DEFAULT_MODELS.LIGHTWEIGHT,
         temperature: 0.3,
     });
 
@@ -180,7 +181,7 @@ async function synthesizeResults(
 
     const chat = new ChatOpenAI({
         openAIApiKey: env.server.OPENAI_API_KEY,
-        modelName: "gpt-5-mini",
+        modelName: DEFAULT_MODELS.LIGHTWEIGHT,
         temperature: 0.2,
     });
 

src/app/api/predictive-document-analysis/services/analysisEngine.ts

@@ -2,10 +2,10 @@ import { ChatOpenAI } from "@langchain/openai";
 import { HumanMessage, SystemMessage } from "@langchain/core/messages";
 import { z } from "zod";
 import { DuckDuckGoSearch } from "@langchain/community/tools/duckduckgo_search";
-import type { 
-    PdfChunk, 
-    AnalysisSpecification, 
-    PredictiveAnalysisResult, 
+import type {
+    PdfChunk,
+    AnalysisSpecification,
+    PredictiveAnalysisResult,
     MissingDocumentPrediction,
     SearchResult
 } from "../types";
@@ -20,6 +20,7 @@ import { document } from "~/server/db/schema";
 import { and, eq, ne } from "drizzle-orm";
 import stringSimilarity from 'string-similarity-js';
 import { ANALYSIS_BATCH_CONFIG } from "~/lib/constants";
+import { DEFAULT_MODELS } from "~/lib/models";
 
 async function withRetry<T>(
     operation: () => Promise<T>,
@@ -149,7 +150,7 @@ export async function callAIAnalysis(
 
     const chat = new ChatOpenAI({
         openAIApiKey: process.env.OPENAI_API_KEY,
-        modelName: "gpt-5.2",
+        modelName: DEFAULT_MODELS.ANALYSIS,
         temperature: 0.3,
     });
 

src/app/api/predictive-document-analysis/services/referenceExtractor.ts

@@ -3,6 +3,7 @@ import { HumanMessage, SystemMessage } from "@langchain/core/messages";
 import { z } from "zod";
 import type { PdfChunk, DocumentReference } from "../types";
 import { groupContentFromChunks, hasSpecificIdentifier } from "../utils/content";
+import { DEFAULT_MODELS } from "~/lib/models";
 
 const ReferenceExtractionSchema = z.object({
     references: z.array(z.object({
@@ -45,7 +46,7 @@ export async function extractReferences(
 
     const chat = new ChatOpenAI({
         openAIApiKey: process.env.OPENAI_API_KEY,
-        modelName: "gpt-5.2",
+        modelName: DEFAULT_MODELS.ANALYSIS,
         temperature: 0.1,
     });
 

src/app/api/predictive-document-analysis/utils/embeddings.ts

@@ -1,4 +1,5 @@
 import { OpenAIEmbeddings } from "@langchain/openai";
+import { DEFAULT_MODELS } from "~/lib/models";
 
 const embeddingCache = new Map<string, number[]>();
 
@@ -10,7 +11,7 @@ export async function getEmbeddings(text: string): Promise<number[]> {
     try {
         const embeddings = new OpenAIEmbeddings({
             openAIApiKey: process.env.OPENAI_API_KEY,
-            modelName: "text-embedding-ada-002",
+            modelName: DEFAULT_MODELS.EMBEDDING,
         });
 
         const [embedding] = await embeddings.embedDocuments([text]);
@@ -30,7 +31,7 @@ export async function batchGetEmbeddings(texts: string[]): Promise<number[][]> {
     try {
         const embeddings = new OpenAIEmbeddings({
             openAIApiKey: process.env.OPENAI_API_KEY,
-            modelName: "text-embedding-ada-002",
+            modelName: DEFAULT_MODELS.EMBEDDING,
         });
 
         const results = await embeddings.embedDocuments(uniqueTexts);

src/app/api/study-agent/chat/route.ts

@@ -9,6 +9,7 @@ import {
 import { db } from "../../../../server/db/index";
 import { pdfChunks, document, users } from "../../../../server/db/schema";
 import { eq, inArray } from "drizzle-orm";
+import { DEFAULT_MODELS } from "~/lib/models";
 
 import type { EmotionTag, StudyAgentChatRequest } from "./types";
 import {
@@ -245,7 +246,7 @@ export async function POST(request: Request) {
     // Initialize OpenAI and generate response
     const chat = new ChatOpenAI({
       openAIApiKey: process.env.OPENAI_API_KEY,
-      modelName: "gpt-4o-mini",
+      modelName: DEFAULT_MODELS.LIGHTWEIGHT,
       temperature: 0.7,
       timeout: 30000,
     });

src/app/api/study-agent/speech-to-text/route.ts

@@ -2,6 +2,7 @@ import { NextResponse } from "next/server";
 import { auth } from "@clerk/nextjs/server";
 import OpenAI from "openai";
 import { File } from "formdata-node";
+import { DEFAULT_MODELS } from "~/lib/models";
 
 /**
  * Speech-to-Text API using OpenAI Whisper
@@ -70,7 +71,7 @@ export async function POST(request: Request) {
 
     const transcription = await openai.audio.transcriptions.create({
       file: audioFileForOpenAI as unknown as globalThis.File, // Type assertion for OpenAI SDK compatibility
-      model: "gpt-4o-transcribe",
+      model: DEFAULT_MODELS.TRANSCRIPTION,
       response_format: "text",
     });
 

src/app/api/uploadDocument/route.ts

@@ -17,6 +17,7 @@ import type { Document } from "langchain/document";
 import { processPDFWithOCR } from "../services/ocrService";
 import { withRateLimit } from "~/lib/rate-limit-middleware";
 import { RateLimitPresets } from "~/lib/rate-limiter";
+import { DEFAULT_MODELS } from "~/lib/models";
 
 interface PDFMetadata {
     loc?: {
@@ -158,7 +159,7 @@ export async function POST(request: Request) {
         }
 
         const embeddings = new OpenAIEmbeddings({
-            model: "text-embedding-ada-002",
+            model: DEFAULT_MODELS.EMBEDDING,
             openAIApiKey: process.env.OPENAI_API_KEY,
         });
 

src/lib/models.ts

@@ -0,0 +1,75 @@
+/**
+ * Centralized AI Model Configuration
+ *
+ * This file contains all OpenAI model identifiers used throughout the application.
+ * Centralizing these constants makes it easier to:
+ * - Update model versions across the codebase
+ * - Maintain consistency in model usage
+ * - Track which models are used where
+ *
+ * @see https://platform.openai.com/docs/models
+ */
+
+/**
+ * Chat completion models for conversational AI tasks
+ */
+export const CHAT_MODELS = {
+  /**
+   * GPT-4o - Most capable model for complex reasoning tasks
+   * Best for: Document analysis, complex Q&A, structured output generation
+   */
+  GPT_4O: "gpt-4o",
+
+  /**
+   * GPT-4o-mini - Fast and cost-effective for simpler tasks
+   * Best for: Query refinement, quick classifications, lightweight processing
+   */
+  GPT_4O_MINI: "gpt-4o-mini",
+} as const;
+
+/**
+ * Embedding models for vector search and similarity
+ */
+export const EMBEDDING_MODELS = {
+  /**
+   * Ada 002 - Standard embedding model for text similarity
+   * Dimensions: 1536
+   * Best for: Document chunking, semantic search, RAG applications
+   */
+  ADA_002: "text-embedding-ada-002",
+} as const;
+
+/**
+ * Audio transcription models for speech-to-text
+ */
+export const TRANSCRIPTION_MODELS = {
+  /**
+   * GPT-4o Transcribe - High-quality speech-to-text transcription
+   * Best for: Audio transcription, voice input processing
+   */
+  GPT_4O_TRANSCRIBE: "gpt-4o-transcribe",
+} as const;
+
+/**
+ * Default model selections for different use cases
+ */
+export const DEFAULT_MODELS = {
+  /** Default model for document analysis and complex reasoning */
+  ANALYSIS: CHAT_MODELS.GPT_4O,
+
+  /** Default model for lightweight tasks like query refinement */
+  LIGHTWEIGHT: CHAT_MODELS.GPT_4O_MINI,
+
+  /** Default model for text embeddings */
+  EMBEDDING: EMBEDDING_MODELS.ADA_002,
+
+  /** Default model for audio transcription */
+  TRANSCRIPTION: TRANSCRIPTION_MODELS.GPT_4O_TRANSCRIBE,
+} as const;
+
+// Type exports for TypeScript usage
+export type ChatModel = (typeof CHAT_MODELS)[keyof typeof CHAT_MODELS];
+export type EmbeddingModel =
+  (typeof EMBEDDING_MODELS)[keyof typeof EMBEDDING_MODELS];
+export type TranscriptionModel =
+  (typeof TRANSCRIPTION_MODELS)[keyof typeof TRANSCRIPTION_MODELS];