Use Jev to classify and split multi-doc PDFs

curl -sS https://router.flatkey.ai/api/alpha/decisions \
  -H "Authorization: Bearer $FLATKEY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "typesafe/jev-1.13",
  "state": "Source: inbox PDF upload (LiteParse excerpt)\nFile: scan_batch_1042.pdf pages 3–5\nOptional: LlamaParse fallback unused\n\n--- page 3 (end of current region) ---\nACME Supplies LLC\nINVOICE #INV-88421\nBill To: Northwind Labs\nItem: Widget kit x12    $480.00\nTax: $38.40\nTotal due: $518.40\nPayment terms: Net 30\nThank you for your business.\nPage 3 of 3\n\n--- page 4 (start of next region) ---\nJORDAN LEE\nSoftware Engineer\njordan.lee@example.com | San Francisco, CA\nEXPERIENCE\nStaff Engineer — Contoso (2021–present)\n- Built typed decision gates for document pipelines\nEDUCATION\nB.S. Computer Science — State University\nSKILLS: TypeScript, Python, PDF tooling\n\nTask: classify the primary doc in this seam window and decide keep_whole / split_at_page / needs_review.\nPolicy: auto-split only at high confidence; never invent a second document from weak OCR.",
  "questions": {
    "doc_category": {
      "type": "choice",
      "instructions": "Classify the primary document type represented by this parsed excerpt for inbox routing. Prefer the document that dominates the excerpt; if two types clearly collide at a seam, still pick the best single label and rely on boundary for the split.",
      "criteria": {
        "invoice": "Bill or invoice with vendor, amounts, line items, or payment terms",
        "contract": "Agreement, MSA, SOW, or signed terms with parties and obligations",
        "resume": "CV or resume with experience, education, or skills sections",
        "receipt": "Payment receipt, expense slip, or short proof-of-purchase",
        "other": "Correspondence or a document that does not fit the labels above"
      }
    },
    "boundary": {
      "type": "choice",
      "instructions": "Decide whether this page or region is a multi-document boundary that should split the PDF. Look for abrupt topic, layout, letterhead, or pagination resets between adjacent excerpts.",
      "criteria": {
        "keep_whole": "The excerpt is continuous with one document; do not split here",
        "split_at_page": "A clear new document starts at this page/region; split here",
        "needs_review": "Signals conflict or OCR is too weak; a human should confirm before splitting"
      }
    },
    "needs_human": {
      "type": "noul",
      "instructions": "A human should confirm classification or the split before downstream routing or file mutation."
    }
  }
}'

// Node.js 18+ (built-in fetch). Run: FLATKEY_API_KEY=... node jev.mjs
const body = {
  "model": "typesafe/jev-1.13",
  "state": "Source: inbox PDF upload (LiteParse excerpt)\nFile: scan_batch_1042.pdf pages 3–5\nOptional: LlamaParse fallback unused\n\n--- page 3 (end of current region) ---\nACME Supplies LLC\nINVOICE #INV-88421\nBill To: Northwind Labs\nItem: Widget kit x12    $480.00\nTax: $38.40\nTotal due: $518.40\nPayment terms: Net 30\nThank you for your business.\nPage 3 of 3\n\n--- page 4 (start of next region) ---\nJORDAN LEE\nSoftware Engineer\njordan.lee@example.com | San Francisco, CA\nEXPERIENCE\nStaff Engineer — Contoso (2021–present)\n- Built typed decision gates for document pipelines\nEDUCATION\nB.S. Computer Science — State University\nSKILLS: TypeScript, Python, PDF tooling\n\nTask: classify the primary doc in this seam window and decide keep_whole / split_at_page / needs_review.\nPolicy: auto-split only at high confidence; never invent a second document from weak OCR.",
  "questions": {
    "doc_category": {
      "type": "choice",
      "instructions": "Classify the primary document type represented by this parsed excerpt for inbox routing. Prefer the document that dominates the excerpt; if two types clearly collide at a seam, still pick the best single label and rely on boundary for the split.",
      "criteria": {
        "invoice": "Bill or invoice with vendor, amounts, line items, or payment terms",
        "contract": "Agreement, MSA, SOW, or signed terms with parties and obligations",
        "resume": "CV or resume with experience, education, or skills sections",
        "receipt": "Payment receipt, expense slip, or short proof-of-purchase",
        "other": "Correspondence or a document that does not fit the labels above"
      }
    },
    "boundary": {
      "type": "choice",
      "instructions": "Decide whether this page or region is a multi-document boundary that should split the PDF. Look for abrupt topic, layout, letterhead, or pagination resets between adjacent excerpts.",
      "criteria": {
        "keep_whole": "The excerpt is continuous with one document; do not split here",
        "split_at_page": "A clear new document starts at this page/region; split here",
        "needs_review": "Signals conflict or OCR is too weak; a human should confirm before splitting"
      }
    },
    "needs_human": {
      "type": "noul",
      "instructions": "A human should confirm classification or the split before downstream routing or file mutation."
    }
  }
};

const res = await fetch("https://router.flatkey.ai/api/alpha/decisions", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.FLATKEY_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify(body),
});
if (!res.ok) throw new Error(`Jev request failed: ${res.status} ${await res.text()}`);
const { answers, usage } = await res.json();
console.log(JSON.stringify(answers, null, 2), usage);

# Python 3.8+ (pip install requests). Run: FLATKEY_API_KEY=... python jev.py
import json, os, requests

body = json.loads(r"""{
  "model": "typesafe/jev-1.13",
  "state": "Source: inbox PDF upload (LiteParse excerpt)\nFile: scan_batch_1042.pdf pages 3–5\nOptional: LlamaParse fallback unused\n\n--- page 3 (end of current region) ---\nACME Supplies LLC\nINVOICE #INV-88421\nBill To: Northwind Labs\nItem: Widget kit x12    $480.00\nTax: $38.40\nTotal due: $518.40\nPayment terms: Net 30\nThank you for your business.\nPage 3 of 3\n\n--- page 4 (start of next region) ---\nJORDAN LEE\nSoftware Engineer\njordan.lee@example.com | San Francisco, CA\nEXPERIENCE\nStaff Engineer — Contoso (2021–present)\n- Built typed decision gates for document pipelines\nEDUCATION\nB.S. Computer Science — State University\nSKILLS: TypeScript, Python, PDF tooling\n\nTask: classify the primary doc in this seam window and decide keep_whole / split_at_page / needs_review.\nPolicy: auto-split only at high confidence; never invent a second document from weak OCR.",
  "questions": {
    "doc_category": {
      "type": "choice",
      "instructions": "Classify the primary document type represented by this parsed excerpt for inbox routing. Prefer the document that dominates the excerpt; if two types clearly collide at a seam, still pick the best single label and rely on boundary for the split.",
      "criteria": {
        "invoice": "Bill or invoice with vendor, amounts, line items, or payment terms",
        "contract": "Agreement, MSA, SOW, or signed terms with parties and obligations",
        "resume": "CV or resume with experience, education, or skills sections",
        "receipt": "Payment receipt, expense slip, or short proof-of-purchase",
        "other": "Correspondence or a document that does not fit the labels above"
      }
    },
    "boundary": {
      "type": "choice",
      "instructions": "Decide whether this page or region is a multi-document boundary that should split the PDF. Look for abrupt topic, layout, letterhead, or pagination resets between adjacent excerpts.",
      "criteria": {
        "keep_whole": "The excerpt is continuous with one document; do not split here",
        "split_at_page": "A clear new document starts at this page/region; split here",
        "needs_review": "Signals conflict or OCR is too weak; a human should confirm before splitting"
      }
    },
    "needs_human": {
      "type": "noul",
      "instructions": "A human should confirm classification or the split before downstream routing or file mutation."
    }
  }
}""")

res = requests.post(
    "https://router.flatkey.ai/api/alpha/decisions",
    headers={"Authorization": f"Bearer {os.environ['FLATKEY_API_KEY']}"},
    json=body,
    timeout=60,
)
res.raise_for_status()
data = res.json()
print(json.dumps(data["answers"], indent=2), data.get("usage"))

No Flatkey key yet? Get one for Jev 1.13 →