Using @edwinfom/resume-intel with Claude 3.5 Haiku and Claude 3.5 Sonnet.

Anthropic

Setup

npm install @ai-sdk/anthropic
import { parseResume } from '@edwinfom/resume-intel'
import { createAnthropic } from '@ai-sdk/anthropic'
 
const model = createAnthropic({
  apiKey: process.env.ANTHROPIC_API_KEY,
})('claude-3-5-haiku-20241022')
 
const result = await parseResume(pdfBuffer, { model })
Model Speed Cost Best for
claude-3-5-haiku-20241022 Fast Low Most CVs
claude-3-5-sonnet-20241022 Medium Higher Complex, non-standard formats

Claude 3.5 Haiku is particularly strong on CVs with unusual formatting or non-standard section names. It follows instructions precisely and rarely hallucinates field values.

Note on PDF vision

Claude 3.5 Sonnet and Haiku support native PDF vision (submitting PDFs directly to the API). However, resume-intel does not use this feature — it always extracts text first and sends clean text to the model.

This is intentional: the spatial reconstruction step (which fixes multi-column layouts) requires text-level processing. Submitting the raw PDF to Claude's vision API would bypass this critical step and produce worse results on complex CV layouts.