Error Handling

Error classes

`ResumeExtractionError`

Thrown when the LLM fails to produce valid structured data after all retry attempts.

When it occurs:

The LLM consistently produces invalid JSON that cannot be repaired
The LLM produces valid JSON that fails Zod validation on every attempt
The AI provider API returns an error (rate limit, network failure, etc.)

import { parseResume, ResumeExtractionError } from '@edwinfom/resume-intel'
 
try {
  const result = await parseResume(pdfBuffer, { model })
} catch (error) {
  if (error instanceof ResumeExtractionError) {
    console.error('Extraction failed:', error.message)
    // "Failed to extract valid resume data after 4 attempts. Last error: ..."
    
    if (error.cause) {
      console.error('Root cause:', error.cause.message)
    }
  }
}

How to reduce occurrence:

Use a more capable model (e.g., gpt-4o instead of gpt-4o-mini)
Increase maxRetries (default: 3)
Add a systemPromptPrefix with domain-specific instructions

`OcrNotEnabledError`

Thrown when a scanned PDF is detected but OCR processing fails.

When it occurs:

tesseract.js encounters an internal error
The PDF pages cannot be rasterized (corrupted PDF)

import { parseResume, OcrNotEnabledError } from '@edwinfom/resume-intel'
 
try {
  const result = await parseResume(pdfBuffer, { model })
} catch (error) {
  if (error instanceof OcrNotEnabledError) {
    console.error('OCR failed:', error.message)
    // Suggest the user provide a text-native PDF
  }
}

Partial results

In task decomposition mode, section failures are non-fatal. If a section fails after all retries, it is omitted from the result rather than throwing an error. The other sections are still returned.

const result = await parseResume(pdfBuffer, { model })
 
// Check which sections succeeded
for (const section of result.meta.sectionResults ?? []) {
  if (!section.success) {
    console.warn(`Section '${section.section}' failed: ${section.error}`)
  }
}
 
// result.data still contains all successfully extracted sections
console.log(result.data.basics?.name) // available even if 'languages' failed

Cancellation

Use an AbortSignal to cancel in-flight requests:

const controller = new AbortController()
 
// Cancel after 30 seconds
setTimeout(() => controller.abort(), 30_000)
 
try {
  const result = await parseResume(pdfBuffer, {
    model,
    abortSignal: controller.signal,
  })
} catch (error) {
  if (error.name === 'AbortError') {
    console.log('Extraction was cancelled')
  }
}

Provider errors

Provider-specific errors (rate limits, authentication failures, network errors) are thrown as-is from the Vercel AI SDK. They are not wrapped in ResumeExtractionError.

try {
  const result = await parseResume(pdfBuffer, { model })
} catch (error) {
  if (error instanceof ResumeExtractionError) {
    // resume-intel validation failure
  } else if (error instanceof OcrNotEnabledError) {
    // OCR failure
  } else {
    // Provider API error — check error.message for details
    console.error('Provider error:', error.message)
  }
}