fix: resolve multi-page PDF crashes and Gemini API ingestion errors (#2781)

* fix: resolve multi-page PDF crashes and Gemini API ingestion errors

* fix type errors

* refactor: address PR review feedback and prevent call stack recursion in regex payload scanner

---------

Co-authored-by: Ofir <ofir@example.com>
Co-authored-by: Amruth Pillai <im.amruth@gmail.com>

package.json

@@ -62,12 +62,12 @@
 		"@tanstack/react-router-ssr-query": "^1.166.2",
 		"@tanstack/react-start": "^1.166.2",
 		"@tanstack/zod-adapter": "^1.166.2",
-		"@tiptap/extension-highlight": "^3.20.0",
-		"@tiptap/extension-table": "^3.20.0",
-		"@tiptap/extension-text-align": "^3.20.0",
-		"@tiptap/pm": "^3.20.0",
-		"@tiptap/react": "^3.20.0",
-		"@tiptap/starter-kit": "^3.20.0",
+		"@tiptap/extension-highlight": "^3.20.1",
+		"@tiptap/extension-table": "^3.20.1",
+		"@tiptap/extension-text-align": "^3.20.1",
+		"@tiptap/pm": "^3.20.1",
+		"@tiptap/react": "^3.20.1",
+		"@tiptap/starter-kit": "^3.20.1",
 		"ai": "^6.0.116",
 		"ai-sdk-ollama": "^3.8.0",
 		"bcrypt": "^6.0.0",
@@ -85,9 +85,12 @@
 		"immer": "^11.1.4",
 		"input-otp": "^1.4.2",
 		"js-cookie": "^3.0.5",
+		"jsonrepair": "^3.13.2",
+		"mammoth": "^1.11.0",
 		"monaco-editor": "^0.55.1",
 		"motion": "^12.35.0",
 		"nodemailer": "^8.0.1",
+		"pdf-parse": "^2.4.5",
 		"pg": "^8.20.0",
 		"puppeteer-core": "^24.38.0",
 		"qrcode.react": "^4.2.0",
@@ -156,7 +159,8 @@
 		"onlyBuiltDependencies": [
 			"bcrypt",
 			"esbuild",
-			"sharp"
+			"sharp",
+			"pdf-parse"
 		]
 	}
 }

src/dialogs/resume/import.tsx

@@ -3,6 +3,7 @@ import { t } from "@lingui/core/macro";
 import { Trans } from "@lingui/react/macro";
 import { DownloadSimpleIcon, FileIcon, UploadSimpleIcon } from "@phosphor-icons/react";
 import { useMutation } from "@tanstack/react-query";
+import { useNavigate } from "@tanstack/react-router";
 import { useEffect, useRef, useState } from "react";
 import { useForm, useWatch } from "react-hook-form";
 import { toast } from "sonner";
@@ -63,7 +64,21 @@ const formSchema = z.discriminatedUnion("type", [
 
 type FormValues = z.infer<typeof formSchema>;
 
+function fileToBase64(file: File): Promise<string> {
+	return new Promise((resolve, reject) => {
+		const reader = new FileReader();
+		reader.onload = () => {
+			const result = reader.result as string;
+			// remove data URL prefix (e.g., "data:application/pdf;base64," or "data:application/vnd...;base64,")
+			resolve(result.split(",")[1]);
+		};
+		reader.onerror = reject;
+		reader.readAsDataURL(file);
+	});
+}
+
 export function ImportResumeDialog(_: DialogProps<"resume.import">) {
+	const navigate = useNavigate();
 	const { enabled: isAIEnabled, provider, model, apiKey, baseURL } = useAIStore();
 	const closeDialog = useDialogStore((state) => state.closeDialog);
 
@@ -135,8 +150,8 @@ export function ImportResumeDialog(_: DialogProps<"resume.import">) {
 				if (!isAIEnabled)
 					throw new Error(t`This feature requires AI Integration to be enabled. Please enable it in the settings.`);
 
-				const arrayBuffer = await values.file.arrayBuffer();
-				const base64 = btoa(String.fromCharCode(...new Uint8Array(arrayBuffer)));
+				// const arrayBuffer = await values.file.arrayBuffer();
+				const base64 = await fileToBase64(values.file);
 
 				data = await client.ai.parsePdf({
 					provider,
@@ -151,8 +166,9 @@ export function ImportResumeDialog(_: DialogProps<"resume.import">) {
 				if (!isAIEnabled)
 					throw new Error(t`This feature requires AI Integration to be enabled. Please enable it in the settings.`);
 
-				const arrayBuffer = await values.file.arrayBuffer();
-				const base64 = btoa(String.fromCharCode(...new Uint8Array(arrayBuffer)));
+				// const arrayBuffer = await values.file.arrayBuffer();
+				const base64 = await fileToBase64(values.file);
+
 				const mediaType =
 					values.file.type === "application/msword"
 						? ("application/msword" as const)
@@ -170,9 +186,10 @@ export function ImportResumeDialog(_: DialogProps<"resume.import">) {
 
 			if (!data) throw new Error("No data was returned from the AI provider.");
 
-			await importResume({ data });
+			const id = await importResume({ data });
 			toast.success(t`Your resume has been imported successfully.`, { id: toastId, description: null });
 			closeDialog();
+			navigate({ to: `/builder/$resumeId`, params: { resumeId: id } });
 		} catch (error: unknown) {
 			if (error instanceof Error) {
 				toast.error(error.message, { id: toastId, description: null });

src/integrations/orpc/router/ai.ts

@@ -2,10 +2,10 @@ import { ORPCError } from "@orpc/client";
 import { type } from "@orpc/server";
 import { AISDKError, type UIMessage } from "ai";
 import { OllamaError } from "ai-sdk-ollama";
-import z, { ZodError } from "zod";
+import z, { flattenError, ZodError } from "zod";
 import type { ResumeData } from "@/schema/resume/data";
 import { protectedProcedure } from "../context";
-import { aiCredentialsSchema, aiProviderSchema, aiService, fileInputSchema, formatZodError } from "../services/ai";
+import { aiCredentialsSchema, aiProviderSchema, aiService, fileInputSchema } from "../services/ai";
 
 type AIProvider = z.infer<typeof aiProviderSchema>;
 
@@ -69,6 +69,10 @@ export const aiRouter = {
 				message: "The AI provider returned an error or is unreachable.",
 				status: 502,
 			},
+			BAD_REQUEST: {
+				message: "The AI returned an improperly formatted structure.",
+				status: 400,
+			},
 		})
 		.handler(async ({ input }): Promise<ResumeData> => {
 			try {
@@ -79,7 +83,10 @@ export const aiRouter = {
 				}
 
 				if (error instanceof ZodError) {
-					throw new Error(formatZodError(error));
+					throw new ORPCError("BAD_REQUEST", {
+						message: "Invalid resume data structure",
+						cause: flattenError(error),
+					});
 				}
 				throw error;
 			}
@@ -111,6 +118,10 @@ export const aiRouter = {
 				message: "The AI provider returned an error or is unreachable.",
 				status: 502,
 			},
+			BAD_REQUEST: {
+				message: "The AI returned an improperly formatted structure.",
+				status: 400,
+			},
 		})
 		.handler(async ({ input }) => {
 			try {
@@ -121,7 +132,10 @@ export const aiRouter = {
 				}
 
 				if (error instanceof ZodError) {
-					throw new Error(formatZodError(error));
+					throw new ORPCError("BAD_REQUEST", {
+						message: "Invalid resume data structure",
+						cause: flattenError(error),
+					});
 				}
 
 				throw error;

src/integrations/orpc/services/ai.server.ts

@@ -0,0 +1,14 @@
+import { extractRawText } from "mammoth";
+import { PDFParse } from "pdf-parse";
+
+export async function extractPdfText(base64Data: string): Promise<string> {
+	const pdfParse = new PDFParse({ data: base64Data });
+	const data = await pdfParse.getText();
+	return data.text;
+}
+
+export async function extractDocxText(base64Data: string): Promise<string> {
+	const buffer = Buffer.from(base64Data, "base64");
+	const data = await extractRawText({ buffer });
+	return data.value;
+}

src/integrations/orpc/services/ai.ts

@@ -13,9 +13,9 @@ import {
 	type UIMessage,
 } from "ai";
 import { createOllama } from "ai-sdk-ollama";
+import { jsonrepair } from "jsonrepair";
 import { match } from "ts-pattern";
-import type { ZodError } from "zod";
-import z, { flattenError } from "zod";
+import z, { flattenError, ZodError } from "zod";
 import chatSystemPromptTemplate from "@/integrations/ai/prompts/chat-system.md?raw";
 import docxParserSystemPrompt from "@/integrations/ai/prompts/docx-parser-system.md?raw";
 import docxParserUserPrompt from "@/integrations/ai/prompts/docx-parser-user.md?raw";
@@ -28,6 +28,204 @@ import {
 } from "@/integrations/ai/tools/patch-resume";
 import type { ResumeData } from "@/schema/resume/data";
 import { defaultResumeData, resumeDataSchema } from "@/schema/resume/data";
+import { isObject } from "@/utils/sanitize";
+import { extractDocxText, extractPdfText } from "./ai.server";
+
+const aiExtractionTemplate = {
+	...defaultResumeData,
+	basics: {
+		...defaultResumeData.basics,
+		customFields: [{ id: "", icon: "", text: "", link: "" }],
+	},
+	sections: {
+		...defaultResumeData.sections,
+		profiles: {
+			...defaultResumeData.sections.profiles,
+			items: [{ id: "", hidden: false, icon: "", network: "", username: "", website: { url: "", label: "" } }],
+		},
+		experience: {
+			...defaultResumeData.sections.experience,
+			items: [
+				{
+					id: "",
+					hidden: false,
+					company: "",
+					position: "",
+					location: "",
+					period: "",
+					website: { url: "", label: "" },
+					description: "",
+				},
+			],
+		},
+		education: {
+			...defaultResumeData.sections.education,
+			items: [
+				{
+					id: "",
+					hidden: false,
+					school: "",
+					degree: "",
+					area: "",
+					grade: "",
+					location: "",
+					period: "",
+					website: { url: "", label: "" },
+					description: "",
+				},
+			],
+		},
+		projects: {
+			...defaultResumeData.sections.projects,
+			items: [{ id: "", hidden: false, name: "", period: "", website: { url: "", label: "" }, description: "" }],
+		},
+		skills: {
+			...defaultResumeData.sections.skills,
+			items: [{ id: "", hidden: false, icon: "", name: "", proficiency: "", level: 0, keywords: [] }],
+		},
+		languages: {
+			...defaultResumeData.sections.languages,
+			items: [{ id: "", hidden: false, language: "", fluency: "", level: 0 }],
+		},
+		interests: {
+			...defaultResumeData.sections.interests,
+			items: [{ id: "", hidden: false, icon: "", name: "", keywords: [] }],
+		},
+		awards: {
+			...defaultResumeData.sections.awards,
+			items: [
+				{ id: "", hidden: false, title: "", awarder: "", date: "", website: { url: "", label: "" }, description: "" },
+			],
+		},
+		certifications: {
+			...defaultResumeData.sections.certifications,
+			items: [
+				{ id: "", hidden: false, title: "", issuer: "", date: "", website: { url: "", label: "" }, description: "" },
+			],
+		},
+		publications: {
+			...defaultResumeData.sections.publications,
+			items: [
+				{ id: "", hidden: false, title: "", publisher: "", date: "", website: { url: "", label: "" }, description: "" },
+			],
+		},
+		volunteer: {
+			...defaultResumeData.sections.volunteer,
+			items: [
+				{
+					id: "",
+					hidden: false,
+					organization: "",
+					location: "",
+					period: "",
+					website: { url: "", label: "" },
+					description: "",
+				},
+			],
+		},
+		references: {
+			...defaultResumeData.sections.references,
+			items: [
+				{ id: "", hidden: false, name: "", position: "", website: { url: "", label: "" }, phone: "", description: "" },
+			],
+		},
+	},
+};
+
+/**
+ * Merges two objects recursively, filling in missing properties in the target object
+ * with values from the source object, but does not overwrite existing properties in the target
+ * unless the source provides a defined, non-null value.
+ *
+ * Both target and source must be plain objects (Record<string, unknown>).
+ * This function does not mutate either argument; returns a new object.
+ *
+ * @param target - The object to merge into (existing values take precedence)
+ * @param source - The object providing default values
+ * @returns The merged object
+ */
+function mergeDefaults<T extends Record<string, unknown>, S extends Record<string, unknown>>(
+	target: T,
+	source: S,
+): T & S {
+	if (!isObject(target) || !isObject(source)) {
+		// Use source value if defined (non-null, non-undefined), else fallback to target
+		return (source !== undefined && source !== null ? source : target) as T & S;
+	}
+
+	const output: Record<string, unknown> = { ...target };
+
+	for (const key of Object.keys(source)) {
+		const sourceValue = source[key];
+		if (sourceValue === undefined || sourceValue === null) {
+			continue;
+		}
+		const targetValue = target[key];
+
+		if (isObject(sourceValue) && isObject(targetValue)) {
+			output[key] = mergeDefaults(targetValue as Record<string, unknown>, sourceValue as Record<string, unknown>);
+		} else if (isObject(sourceValue) && (targetValue === undefined || targetValue === null)) {
+			// Fill with source object only if target does not have it
+			output[key] = sourceValue;
+		} else if (!isObject(sourceValue)) {
+			output[key] = sourceValue;
+		} else if (targetValue === undefined) {
+			output[key] = sourceValue;
+		}
+	}
+
+	return output as T & S;
+}
+
+function logAndRethrow(context: string, error: unknown): never {
+	if (error instanceof Error) {
+		console.error(`${context}:`, error.message);
+		throw error;
+	}
+	console.error(`Unknown error in ${context}:`, error);
+	throw new Error(`An unknown error occurred during ${context}.`);
+}
+
+function parseAndValidateResumeJson(resultText: string): ResumeData {
+	let jsonString = resultText;
+	const firstCurly = jsonString.indexOf("{");
+	const firstSquare = jsonString.indexOf("[");
+	const lastCurly = jsonString.lastIndexOf("}");
+	const lastSquare = jsonString.lastIndexOf("]");
+
+	let firstIndex = -1;
+	if (firstCurly !== -1 && firstSquare !== -1) {
+		firstIndex = Math.min(firstCurly, firstSquare);
+	} else {
+		firstIndex = Math.max(firstCurly, firstSquare);
+	}
+	const lastIndex = Math.max(lastCurly, lastSquare);
+
+	if (firstIndex !== -1 && lastIndex !== -1 && lastIndex >= firstIndex) {
+		jsonString = jsonString.substring(firstIndex, lastIndex + 1);
+	}
+
+	try {
+		const repairedJson = jsonrepair(jsonString);
+		const parsedJson = JSON.parse(repairedJson);
+		const mergedData = mergeDefaults(defaultResumeData, parsedJson);
+
+		return resumeDataSchema.parse({
+			...mergedData,
+			customSections: [],
+			picture: defaultResumeData.picture,
+			metadata: defaultResumeData.metadata,
+		});
+	} catch (error: unknown) {
+		if (error instanceof ZodError) {
+			console.error("Zod Validation Errors:", JSON.stringify(flattenError(error), null, 2));
+			throw error;
+		}
+
+		console.error("Unknown error:", error);
+		throw new Error("An unknown error occurred while validating the merged resume data.");
+	}
+}
 
 export const aiProviderSchema = z.enum(["ollama", "openai", "gemini", "anthropic", "vercel-ai-gateway"]);
 
@@ -89,35 +287,28 @@ type ParsePdfInput = z.infer<typeof aiCredentialsSchema> & {
 async function parsePdf(input: ParsePdfInput): Promise<ResumeData> {
 	const model = getModel(input);
 
+	const pdfText = await extractPdfText(input.file.data).catch((error: unknown) =>
+		logAndRethrow("Failed to parse PDF locally", error),
+	);
+
 	const result = await generateText({
 		model,
-		output: Output.object({ schema: resumeDataSchema }),
 		messages: [
 			{
 				role: "system",
-				content: pdfParserSystemPrompt,
+				content:
+					pdfParserSystemPrompt +
+					"\n\nIMPORTANT: You must return ONLY raw valid JSON. Do not return markdown, do not return explanations. Just the JSON object. Use the following JSON as a template and fill in the extracted values. For arrays, you MUST use the exact key names shown in the template (e.g. use 'description' instead of 'summary', 'website' instead of 'url'):\n\n" +
+					JSON.stringify(aiExtractionTemplate, null, 2),
 			},
 			{
 				role: "user",
-				content: [
-					{ type: "text", text: pdfParserUserPrompt },
-					{
-						type: "file",
-						filename: input.file.name,
-						mediaType: "application/pdf",
-						data: input.file.data,
-					},
-				],
+				content: `${pdfParserUserPrompt}\n\n--- EXTRACTED RESUME TEXT ---\n${pdfText}\n--- END OF EXTRACTED TEXT ---`,
 			},
 		],
-	});
+	}).catch((error: unknown) => logAndRethrow("Failed to generate the text with the model", error));
 
-	return resumeDataSchema.parse({
-		...result.output,
-		customSections: [],
-		picture: defaultResumeData.picture,
-		metadata: defaultResumeData.metadata,
-	});
+	return parseAndValidateResumeJson(result.text);
 }
 
 type ParseDocxInput = z.infer<typeof aiCredentialsSchema> & {
@@ -128,36 +319,28 @@ type ParseDocxInput = z.infer<typeof aiCredentialsSchema> & {
 async function parseDocx(input: ParseDocxInput): Promise<ResumeData> {
 	const model = getModel(input);
 
+	const docxText = await extractDocxText(input.file.data).catch((error: unknown) =>
+		logAndRethrow("Failed to parse DOCX locally", error),
+	);
+
 	const result = await generateText({
 		model,
-		output: Output.object({ schema: resumeDataSchema }),
 		messages: [
-			{ role: "system", content: docxParserSystemPrompt },
+			{
+				role: "system",
+				content:
+					docxParserSystemPrompt +
+					"\n\nIMPORTANT: You must return ONLY raw valid JSON. Do not return markdown, do not return explanations. Just the JSON object. Use the following JSON as a template and fill in the extracted values. For arrays, you MUST use the exact key names shown in the template (e.g. use 'description' instead of 'summary', 'website' instead of 'url'):\n\n" +
+					JSON.stringify(aiExtractionTemplate, null, 2),
+			},
 			{
 				role: "user",
-				content: [
-					{ type: "text", text: docxParserUserPrompt },
-					{
-						type: "file",
-						filename: input.file.name,
-						mediaType: input.mediaType,
-						data: input.file.data,
-					},
-				],
+				content: `${docxParserUserPrompt}\n\n--- EXTRACTED RESUME TEXT ---\n${docxText}\n--- END OF EXTRACTED TEXT ---`,
 			},
 		],
-	});
+	}).catch((error: unknown) => logAndRethrow("Failed to generate the text with the model", error));
 
-	return resumeDataSchema.parse({
-		...result.output,
-		customSections: [],
-		picture: defaultResumeData.picture,
-		metadata: defaultResumeData.metadata,
-	});
-}
-
-export function formatZodError(error: ZodError): string {
-	return JSON.stringify(flattenError(error));
+	return parseAndValidateResumeJson(result.text);
 }
 
 function buildChatSystemPrompt(resumeData: ResumeData): string {

src/routes/builder/$resumeId/-sidebar/right/index.tsx

@@ -47,7 +47,10 @@ export function BuilderSidebarRight() {
 		<>
 			<SidebarEdge scrollAreaRef={scrollAreaRef} />
 
-			<ScrollArea ref={scrollAreaRef} className="@container h-[calc(100svh-3.5rem)] bg-background overflow-hidden sm:me-12">
+			<ScrollArea
+				ref={scrollAreaRef}
+				className="@container h-[calc(100svh-3.5rem)] overflow-hidden bg-background sm:me-12"
+			>
 				<div className="space-y-4 p-4">
 					{rightSidebarSections.map((section) => (
 						<Fragment key={section}>
@@ -85,18 +88,18 @@ function SidebarEdge({ scrollAreaRef }: SidebarEdgeProps) {
 		<BuilderSidebarEdge side="right">
 			<div className="no-scrollbar min-h-0 w-full flex-1 overflow-y-auto overflow-x-hidden">
 				<div className="flex min-h-full flex-col items-center justify-center gap-y-2">
-				{rightSidebarSections.map((section) => (
-					<Button
-						key={section}
-						size="icon"
-						variant="ghost"
-						title={getSectionTitle(section)}
-						onClick={() => scrollToSection(section)}
-					>
-						{getSectionIcon(section)}
-					</Button>
-				))}
-			</div>
+					{rightSidebarSections.map((section) => (
+						<Button
+							key={section}
+							size="icon"
+							variant="ghost"
+							title={getSectionTitle(section)}
+							onClick={() => scrollToSection(section)}
+						>
+							{getSectionIcon(section)}
+						</Button>
+					))}
+				</div>
 			</div>
 		</BuilderSidebarEdge>
 	);

src/utils/sanitize.ts

@@ -109,3 +109,13 @@ export function sanitizeCss(css: string): string {
 
 	return sanitized;
 }
+
+/**
+ * Checks if the given value is a plain JSON object (not null, not array, not other types).
+ *
+ * @param value - The value to check.
+ * @returns True if the value is a plain object, false otherwise.
+ */
+export function isObject(value: unknown): value is Record<string, unknown> {
+	return typeof value === "object" && value !== null && !Array.isArray(value);
+}