Four representative examples of how the four nodes are assembled into real work. If your task is similar, take the structure as is and change only the schema and the prompt.

Resume screening

• Extract — a schema for name, contact, work history, and tech stack.
• Instruct — use Generate and decide, put criteria such as "5+ years of relevant experience preferred; flag for follow-up when information is missing" in the prompt, and decide between Advance / Needs review / Reject.
• Why it works — details scattered differently across every applicant get sorted onto the same axis, giving you comparable evidence.

---

Expense processing

• Classify — define the document types Domestic receipt / Overseas receipt / Other.
• Extract — map a schema per type (domestic: merchant, issue date, amount / overseas: store_name, currency, total_amount).
• Instruct — decide approve or reject on a rule such as "reject if the total exceeds 10,000 KRW".
• Why it works — upload receipts as individual files and the same review runs against every case. When you bundle several receipts into one case, add Split mode as in the next example.

---

Mixed scan bundles — using Split

A stack of scans containing contracts, receipts, and invoices all at once.

• Classify — Split, then classify mode with Auto-detect documents as the split criterion. It finds where each new document starts inside the bundle, cuts there, and classifies each piece.
• Extract — extracts from each split document with the schema mapped to its type, producing as many results as there are documents.
• Why it works — in Split mode one case yields several document-level results, so reviewing row by row in the table view is much easier.

---

Comparing quotes

• Extract — pull item, unit price, total, and delivery date with a table schema.
• Instruct — reference @item and @total_amount to build a vendor comparison table, summarizing the lowest price and the differences in terms.
• Why it works — vendors use different layouts, but Extract normalizes them into the same structure, which makes the comparison straightforward.

---

Contract review

• Classify — sort by contract type (services, purchase, NDA, and so on).
• Extract — pull the term, value, penalty, and termination clauses.
• Instruct — compare against your standard terms, flag unfavorable or ambiguous clauses, and judge whether legal review is needed.
• Why it works — the more you state the criteria as numbers ("needs review if the penalty clause exceeds 10% of the contract value"), the more consistent the outcome.