Extraction API

Pipeline functions

classifyDocumentType(pdf, options?)

Classify a document as a policy or quote.

Parameter	Type	Description
pdf	string	Base64-encoded PDF
options	ClassifyOptions	Optional configuration

Returns: Promise<{ documentType: "policy" | "quote"; confidence: number; signals: string[] }>

const { documentType, confidence, signals } = await classifyDocumentType(pdfBase64);

---

extractFromPdf(pdf, options?)

Full policy extraction pipeline (passes 1-3).

Parameter	Type	Description
pdf	string	Base64-encoded PDF
options	ExtractOptions	Optional configuration

Returns: Promise<{ rawText: string; extracted: any }>

const { rawText, extracted } = await extractFromPdf(pdfBase64, {
  log: async (msg) => console.log(msg),
  onMetadata: async (raw) => await saveMetadata(raw),
  models: customModels,
});

---

extractQuoteFromPdf(pdf, options?)

Full quote extraction pipeline (passes 1-2).

Parameters: Same as extractFromPdf.

Returns: Promise<{ rawText: string; extracted: any }>

---

extractSectionsOnly(pdf, metadataRaw, options?)

Retry section extraction using saved metadata from a prior pass 1.

Parameter	Type	Description
pdf	string	Base64-encoded PDF
metadataRaw	string	JSON string from a previous pass 1
options	ExtractSectionsOptions	Optional configuration

Returns: Promise<{ rawText: string; extracted: any }>

---

enrichSupplementaryFields(document, models?, log?)

Pass 3 enrichment. Parses raw text into structured supplementary fields. Non-fatal — returns the document unchanged on failure.

Parameter	Type	Description
document	any	Document object with raw supplementary fields
models	ModelConfig	Optional model config
log	LogFn	Optional logger

Returns: Promise<any>

Mapping functions

applyExtracted(extracted)

Map raw policy extraction JSON to persistence-ready fields.

Returns: Object with carrier, policyNumber, coverages, effectiveDate, expirationDate, etc.

---

applyExtractedQuote(extracted)

Map raw quote extraction JSON to persistence-ready fields.

Returns: Object with quoteNumber, premiumBreakdown, subjectivities, proposedEffectiveDate, etc.

Merge functions

mergeChunkedSections(metadataResult, sectionChunks)

Merge section chunks from policy extraction. Combines sections and takes the last non-null supplementary field.

---

mergeChunkedQuoteSections(metadataResult, sectionChunks)

Merge section chunks from quote extraction. Also accumulates subjectivities and underwriting conditions.

Utility functions

getPageChunks(totalPages, chunkSize?)

Calculate page ranges for chunked extraction.

Parameter	Type	Default	Description
totalPages	number	—	Total page count
chunkSize	number	30	Pages per chunk

Returns: Array<[number, number]>

---

stripFences(text)

Remove markdown code fences from AI response text.

---

sanitizeNulls<T>(obj)

Recursively convert null values to undefined.

Options types

ExtractOptions

interface ExtractOptions {
  log?: LogFn;
  onMetadata?: (raw: string) => Promise<void>;
  models?: ModelConfig;
  metadataProviderOptions?: ProviderOptions;
  fallbackProviderOptions?: ProviderOptions;
}

ClassifyOptions

interface ClassifyOptions {
  log?: LogFn;
  models?: ModelConfig;
}

ExtractSectionsOptions

interface ExtractSectionsOptions {
  log?: LogFn;
  promptBuilder?: PromptBuilder;
  models?: ModelConfig;
  fallbackProviderOptions?: ProviderOptions;
}

LogFn

type LogFn = (message: string) => Promise<void>;

Constants

Constant	Value	Description
SONNET_MODEL	"claude-sonnet-4-6"	Default Sonnet model ID
HAIKU_MODEL	"claude-haiku-4-5-20251001"	Default Haiku model ID