Auto-fill thesis abstract from uploaded PDFs (#1127)

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: krusche

client/e2e/helpers.ts

@@ -235,3 +235,42 @@ export function createTestPdfBuffer(): Buffer {
       'trailer\n<< /Size 4 /Root 1 0 R >>\nstartxref\n206\n%%EOF',
   )
 }
+
+/**
+ * Create a valid PDF buffer with a real "Abstract" section so the server's abstract
+ * extractor can confidently extract it. The body contains a line-end hyphenation
+ * ("com-" / "prehensive") that must be rejoined into "comprehensive", and a
+ * "1 Introduction" heading that bounds the abstract.
+ */
+export function createAbstractTestPdfBuffer(): Buffer {
+  const content =
+    'BT /F1 14 Tf 72 720 Td (Abstract) Tj ET\n' +
+    'BT /F1 10 Tf 72 695 Td (This thesis presents a com-) Tj ET\n' +
+    'BT /F1 10 Tf 72 680 Td (prehensive evaluation of automated review systems.) Tj ET\n' +
+    'BT /F1 14 Tf 72 650 Td (1 Introduction) Tj ET\n' +
+    'BT /F1 10 Tf 72 625 Td (The introduction begins here with more detail.) Tj ET\n'
+
+  const objects = [
+    '<< /Type /Catalog /Pages 2 0 R >>',
+    '<< /Type /Pages /Kids [3 0 R] /Count 1 >>',
+    '<< /Type /Page /Parent 2 0 R /MediaBox [0 0 612 792] /Resources << /Font << /F1 4 0 R >> >> /Contents 5 0 R >>',
+    '<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>',
+    `<< /Length ${Buffer.byteLength(content)} >>\nstream\n${content}endstream`,
+  ]
+
+  let pdf = '%PDF-1.4\n'
+  const offsets: number[] = []
+  objects.forEach((body, index) => {
+    offsets.push(Buffer.byteLength(pdf))
+    pdf += `${index + 1} 0 obj\n${body}\nendobj\n`
+  })
+
+  const xrefOffset = Buffer.byteLength(pdf)
+  pdf += `xref\n0 ${objects.length + 1}\n0000000000 65535 f \n`
+  offsets.forEach((offset) => {
+    pdf += `${String(offset).padStart(10, '0')} 00000 n \n`
+  })
+  pdf += `trailer\n<< /Size ${objects.length + 1} /Root 1 0 R >>\nstartxref\n${xrefOffset}\n%%EOF`
+
+  return Buffer.from(pdf, 'latin1')
+}

client/e2e/thesis/abstract-extraction.spec.ts

@@ -0,0 +1,85 @@
+import { test, expect, type Page } from '@playwright/test'
+import {
+  authStatePath,
+  navigateTo,
+  expandAccordion,
+  hideWebpackOverlay,
+  createAbstractTestPdfBuffer,
+} from '../helpers'
+
+// Seeded thesis in WRITING state owned by the "student" user (same thesis the
+// thesis file upload test uses). Uploading a thesis document triggers the
+// server-side abstract extraction.
+const THESIS_1_ID = '00000000-0000-4000-d000-000000000001'
+const THESIS_1_URL = `/theses/${THESIS_1_ID}`
+
+const EXTRACTED = /comprehensive evaluation of automated review systems/i
+
+async function uploadThesisPdf(page: Page) {
+  await expandAccordion(page, 'Thesis', page.getByText('Files').first())
+  await expandAccordion(page, 'Files', page.getByRole('button', { name: 'Upload Thesis' }))
+
+  const uploadThesisButton = page.getByRole('button', { name: 'Upload Thesis' })
+  await uploadThesisButton.scrollIntoViewIfNeeded()
+  await uploadThesisButton.click()
+
+  // Scope to the upload dialog by name: after upload the confirmation modal opens, so a bare
+  // getByRole('dialog') would match two dialogs.
+  const dialog = page.getByRole('dialog', { name: 'File Upload' })
+  await expect(dialog).toBeVisible({ timeout: 10_000 })
+  await dialog.locator('input[type="file"]').setInputFiles({
+    name: 'thesis-with-abstract.pdf',
+    mimeType: 'application/pdf',
+    buffer: createAbstractTestPdfBuffer(),
+  })
+
+  const uploadFileButton = dialog.getByRole('button', { name: 'Upload File' })
+  await expect(uploadFileButton).toBeEnabled({ timeout: 10_000 })
+  await uploadFileButton.click()
+  await expect(dialog).toBeHidden({ timeout: 10_000 })
+}
+
+test.describe('Abstract auto-extraction - Student', () => {
+  test.use({ storageState: authStatePath('student') })
+
+  test('auto-fills a blank abstract, then asks via a modal before replacing an existing one', async ({
+    page,
+  }) => {
+    test.setTimeout(120_000)
+
+    await navigateTo(page, THESIS_1_URL)
+    await hideWebpackOverlay(page)
+    await expect(page.getByRole('heading', { name: /automated code review/i })).toBeVisible({
+      timeout: 30_000,
+    })
+
+    // 1) The seeded abstract is blank, so a confident extraction fills it silently — no modal.
+    //    The line-end hyphenation is rejoined ("com-" + "prehensive" => "comprehensive").
+    await uploadThesisPdf(page)
+    await expect(page.getByText(EXTRACTED)).toBeVisible({ timeout: 15_000 })
+
+    // 2) Replace the abstract with a manual edit; a later upload must NOT overwrite it silently.
+    await page.getByRole('button', { name: 'Edit' }).first().click()
+    const abstractEditor = page
+      .locator('.mantine-InputWrapper-root', { hasText: 'Abstract' })
+      .locator('.ProseMirror')
+      .first()
+    await abstractEditor.click()
+    await page.keyboard.press('ControlOrMeta+A')
+    await page.keyboard.press('Backspace')
+    await abstractEditor.pressSequentially('A manually written abstract that must not be overwritten.')
+    await page.getByRole('button', { name: 'Save' }).click()
+    await expect(page.getByText(/manually written abstract that must not be overwritten/i)).toBeVisible({
+      timeout: 10_000,
+    })
+
+    // 3) Uploading again surfaces a confirmation modal instead of overwriting the manual abstract.
+    await uploadThesisPdf(page)
+    const modal = page.getByRole('dialog', { name: /use the abstract extracted/i })
+    await expect(modal).toBeVisible({ timeout: 15_000 })
+
+    // 4) Confirming replaces the abstract with the extracted text.
+    await modal.getByRole('button', { name: 'Use extracted abstract' }).click()
+    await expect(page.getByText(EXTRACTED)).toBeVisible({ timeout: 10_000 })
+  })
+})

client/src/thesis/pages/ThesisPage/components/ThesisInfoSection/ThesisInfoSection.tsx

@@ -10,6 +10,7 @@ import {
 import { Link } from 'react-router'
 import { ApiError } from '@/core/requests/handler'
 import DownloadAllFilesButton from '@/thesis/pages/ThesisPage/components/ThesisInfoSection/components/DownloadAllFilesButton/DownloadAllFilesButton'
+import AbstractSuggestionModal from '@/thesis/pages/ThesisPage/components/ThesisInfoSection/components/AbstractSuggestionModal/AbstractSuggestionModal'
 import { isThesisClosed } from '@/thesis/utils/thesis'
 import { GLOBAL_CONFIG } from '@/core/config/global'
 import { formatLanguage } from '@/core/utils/format'
@@ -54,6 +55,35 @@ const ThesisInfoSection = () => {
     }
   }, 'Thesis info updated successfully')
 
+  const [accepting, onUseSuggestion] = useThesisUpdateAction(async () => {
+    const response = await doRequest<IThesis>(
+      `/v2/theses/${thesis.thesisId}/abstract-suggestion/accept`,
+      { method: 'POST', requiresAuth: true },
+    )
+
+    if (response.ok) {
+      return response.data
+    } else {
+      throw new ApiError(response)
+    }
+  }, 'Abstract updated from the suggestion')
+
+  const [dismissing, onDismissSuggestion] = useThesisUpdateAction(async () => {
+    const response = await doRequest<IThesis>(
+      `/v2/theses/${thesis.thesisId}/abstract-suggestion/dismiss`,
+      { method: 'POST', requiresAuth: true },
+    )
+
+    if (response.ok) {
+      return response.data
+    } else {
+      throw new ApiError(response)
+    }
+  }, 'Abstract suggestion dismissed')
+
+  const showSuggestion =
+    access.student && !editMode && !isThesisClosed(thesis) && !!thesis.abstractSuggestion
+
   return (
     <Accordion variant='separated' defaultValue='open'>
       <Accordion.Item value='open'>
@@ -83,6 +113,14 @@ const ThesisInfoSection = () => {
                 ))}
               </Stack>
             )}
+            <AbstractSuggestionModal
+              opened={showSuggestion}
+              currentAbstract={thesis.abstractText ?? ''}
+              suggestion={thesis.abstractSuggestion ?? ''}
+              loading={accepting || dismissing}
+              onConfirm={onUseSuggestion}
+              onDeny={onDismissSuggestion}
+            />
             <DocumentEditor
               label='Abstract'
               value={abstractText}

client/src/thesis/pages/ThesisPage/components/ThesisInfoSection/components/AbstractSuggestionModal/AbstractSuggestionModal.tsx

@@ -0,0 +1,74 @@
+import { Button, Group, Modal, Stack, Text } from '@mantine/core'
+import { Sparkle } from '@phosphor-icons/react'
+import DocumentEditor from '@/core/components/DocumentEditor/DocumentEditor'
+
+interface IAbstractSuggestionModalProps {
+  opened: boolean
+  currentAbstract: string
+  suggestion: string
+  loading: boolean
+  onConfirm: () => void
+  onDeny: () => void
+}
+
+// Detect visible text by parsing the HTML and reading its text content, rather than stripping
+// tags with a regex (which is incomplete and flagged as unsafe sanitization). The parsed document
+// is detached and inert, so no scripts run.
+const hasText = (html: string) =>
+  (new DOMParser().parseFromString(html, 'text/html').body.textContent ?? '').trim().length > 0
+
+/**
+ * Shown right after an upload when an abstract was extracted that would replace the current one,
+ * so the student explicitly confirms or denies the change rather than overlooking an inline hint.
+ * Closing the modal denies the change and keeps the current abstract.
+ */
+const AbstractSuggestionModal = (props: IAbstractSuggestionModalProps) => {
+  const { opened, currentAbstract, suggestion, loading, onConfirm, onDeny } = props
+
+  return (
+    <Modal
+      opened={opened}
+      onClose={() => {
+        // Don't let a close-based deny fire while an accept/deny request is already in flight.
+        if (!loading) {
+          onDeny()
+        }
+      }}
+      size='lg'
+      closeOnClickOutside={false}
+      closeOnEscape={!loading}
+      withCloseButton={!loading}
+      title={
+        <Group gap='xs'>
+          <Sparkle size={18} />
+          <Text fw={600}>Use the abstract extracted from your upload?</Text>
+        </Group>
+      }
+    >
+      <Stack>
+        <Text size='sm'>
+          We extracted this abstract from the document you just uploaded. Replace the current
+          abstract with it, or keep the one you have.
+        </Text>
+        <DocumentEditor label='Extracted from your upload' value={suggestion} editMode={false} />
+        {hasText(currentAbstract) && (
+          <DocumentEditor
+            label='Current abstract (kept if you decline)'
+            value={currentAbstract}
+            editMode={false}
+          />
+        )}
+        <Group justify='flex-end'>
+          <Button variant='default' color='gray' loading={loading} onClick={onDeny}>
+            Keep current
+          </Button>
+          <Button loading={loading} onClick={onConfirm}>
+            Use extracted abstract
+          </Button>
+        </Group>
+      </Stack>
+    </Modal>
+  )
+}
+
+export default AbstractSuggestionModal

client/src/thesis/requests/responses/thesis.ts

@@ -65,6 +65,8 @@ export interface IThesis extends IThesisOverview {
   visibility: string
   infoText?: string
   abstractText?: string
+  abstractSource?: 'MANUAL' | 'EXTRACTED'
+  abstractSuggestion?: string
   applicationId: string | null
   anonymized?: boolean
   anonymizedAt?: string