chore(worker): restructure inference modules and centralize config

- Remove legacy inference and platform files from `src/inference/` and migrate responsibilities to new modular locations under `src/infrastructure/`, `src/inference/pdf/`, and `src/inference/whisper/`
- Consolidate environment/config logic into `src/infrastructure/config.ts`
- Move docstore and ffmpeg platform utilities to `src/infrastructure/platform.ts`
- Refactor PDF and Whisper inference code to use new document layout, layout model, and timestamp utilities
- Update API contracts to define and export types and constants previously scattered in inference/types
- Adjust all imports in jobs, storage, and tests to reference new module structure and type locations
- Inline parser version and encoding logic into API contracts
- Remove obsolete files and update test fixtures for new type locations

This update improves codebase modularity, maintainability, and separation of concerns. No changes to inference or orchestration functionality.

BREAKING CHANGE: inference module structure, type imports, and config utilities have changed; downstream code must update imports and integration points.

compute-worker/src/api/app.ts

@@ -15,13 +15,13 @@ import {
 import { Kvm } from '@nats-io/kv';
 import {
   ensureComputeModels,
-} from '../inference/local-runtime';
+} from '../inference/runtime';
 import {
   getComputeTimeoutConfig,
   getComputeOpStaleMs,
   getAvailableCpuCores,
   getOnnxThreadsPerJob,
-} from '../inference';
+} from '../infrastructure/config';
 import { encodeSseFrame, OperationOrchestrator } from '../operations';
 import type {
   PdfLayoutJobRequest,

compute-worker/src/api/contracts.ts

@@ -1,24 +1,26 @@
-import type { TTSSentenceAlignment } from '../inference/types/tts';
+import type { TTSSentenceAlignment, ParsedPdfDocument } from './types';
-import type { ParsedPdfDocument } from '../inference/types/parsed-pdf';
 
 export type {
   TTSAudioBuffer,
   TTSAudioBytes,
   TTSSentenceAlignment,
   TTSSentenceWord,
-} from '../inference/types/tts';
+} from './types';
 export type {
   ParsedPdfBlockKind,
   ParsedPdfBlockFragment,
   ParsedPdfBlock,
   ParsedPdfPage,
   ParsedPdfDocument,
-} from '../inference/types/parsed-pdf';
+} from './types';
 
 export const ALIGN_QUEUE_NAME = 'whisper-align';
 export const PDF_LAYOUT_QUEUE_NAME = 'pdf-layout';
-export { PDF_PARSER_VERSION } from '../inference/pdf/parser-version';
+export const PDF_PARSER_VERSION = 'pp-doclayoutv3-onnx@800+pdfjs@4.8.69';
-export { encodeParserVersion } from '../inference/pdf/parser-version-key';
+
+export function encodeParserVersion(parserVersion: string, defaultVersion = PDF_PARSER_VERSION): string {
+  return encodeURIComponent(parserVersion.trim() || defaultVersion);
+}
 
 export interface WhisperAlignJobBase {
   text: string;

compute-worker/src/api/operation-keys.ts

@@ -1,5 +1,5 @@
 import { createHash } from 'node:crypto';
-import { PDF_PARSER_VERSION } from '../inference/pdf/parser-version';
+import { PDF_PARSER_VERSION } from '../api/contracts';
 
 function sha256Hex(input: string): string {
   return createHash('sha256').update(input).digest('hex');

compute-worker/src/api/types.ts

@@ -61,3 +61,20 @@ export interface PdfParseProgress {
   currentPage?: number;
   phase: PdfParsePhase;
 }
+
+export type TTSAudioBuffer = ArrayBuffer;
+export type TTSAudioBytes = number[];
+
+export interface TTSSentenceWord {
+  text: string;
+  startSec: number;
+  endSec: number;
+  charStart: number;
+  charEnd: number;
+}
+
+export interface TTSSentenceAlignment {
+  sentence: string;
+  sentenceIndex: number;
+  words: TTSSentenceWord[];
+}

compute-worker/src/inference/config/cpu-budget.ts

@@ -1,28 +0,0 @@
-import os from 'node:os';
-
-function readPositiveInt(name: string, fallback: number): number {
-  const raw = process.env[name]?.trim();
-  if (!raw) return fallback;
-  const parsed = Number(raw);
-  if (!Number.isFinite(parsed) || parsed <= 0) return fallback;
-  return Math.floor(parsed);
-}
-
-export function getComputeJobConcurrency(): number {
-  return readPositiveInt('COMPUTE_JOB_CONCURRENCY', 1);
-}
-
-export function getAvailableCpuCores(): number {
-  if (typeof os.availableParallelism === 'function') {
-    const value = os.availableParallelism();
-    if (Number.isFinite(value) && value >= 1) return Math.floor(value);
-  }
-  const fallback = os.cpus().length;
-  return Number.isFinite(fallback) && fallback >= 1 ? Math.floor(fallback) : 1;
-}
-
-export function getOnnxThreadsPerJob(): number {
-  const concurrency = getComputeJobConcurrency();
-  const usableCores = Math.max(1, getAvailableCpuCores() - 1);
-  return Math.max(1, Math.floor(usableCores / concurrency));
-}

compute-worker/src/inference/index.ts

@@ -1,26 +0,0 @@
-export * from '../api/contracts';
-export {
-  getComputeJobConcurrency,
-  getAvailableCpuCores,
-  getOnnxThreadsPerJob,
-} from './config/cpu-budget';
-export {
-  getComputeTimeoutConfig,
-  getComputeOpStaleMs,
-  getWorkerClientWaitTimeoutMs,
-  withTimeout,
-  withIdleTimeoutAndHardCap,
-  type ComputeTimeoutConfig,
-  type ComputeOperationKind,
-  type IdleTimeoutAndHardCapInput,
-} from './config/timeout';
-export { renderPage } from './pdf/render';
-export { mergeTextWithRegions } from './pdf/merge';
-export { PDF_PARSER_VERSION } from './pdf/parser-version';
-export { encodeParserVersion } from './pdf/parser-version-key';
-export { stitchCrossPageBlocks } from './pdf/stitch';
-export { normalizeTextItemsForLayout } from './pdf/normalize-text';
-export { mapWordsToSentenceOffsets, type WhisperWord } from './whisper/alignment-map';
-export { buildGoertzelCoefficients, goertzelPower } from './whisper/spectral';
-export { buildWordsFromTimestampedTokens, extractTokenStartTimestamps } from './whisper/token-timestamps';
-export * from '../operations';

compute-worker/src/inference/pdf/document-layout.ts

@@ -1,4 +1,18 @@
-import type { LayoutRegion, PdfTextItem } from './types';
+import type { ParsedPdfBlockKind } from '../../api/types';
+
+export interface PdfTextItem {
+  text: string;
+  x: number;
+  y: number;
+  width: number;
+  height: number;
+}
+
+export interface LayoutRegion {
+  bbox: [number, number, number, number];
+  label: ParsedPdfBlockKind;
+  confidence?: number;
+}
 
 const NON_TEXT_REGION_LABELS = new Set<LayoutRegion['label']>(['chart', 'image', 'table', 'seal']);
 const TEXT_ASSIGNABLE_LABELS = new Set<LayoutRegion['label']>([

compute-worker/src/inference/pdf/layout-model.ts

@@ -1,8 +1,8 @@
 import * as ort from 'onnxruntime-node';
 import { readFile } from 'fs/promises';
-import type { LayoutRegion, PdfTextItem } from './types';
+import type { LayoutRegion, PdfTextItem } from './document-layout';
 import { ensureModel, MODEL_CONFIG_PATH, MODEL_PREPROCESSOR_PATH } from './model';
-import { getOnnxThreadsPerJob } from '../config/cpu-budget';
+import { getOnnxThreadsPerJob } from '../../infrastructure/config';
 
 interface RunLayoutInput {
   pageWidth: number;

compute-worker/src/inference/pdf/model.ts

@@ -3,7 +3,7 @@ import { fileURLToPath } from 'url';
 import { createHash } from 'crypto';
 import { readFileSync } from 'fs';
 import { access, mkdir, rename, writeFile, readFile, unlink, copyFile } from 'fs/promises';
-import { DOCSTORE_DIR } from '../platform/docstore';
+import { DOCSTORE_DIR } from '../../infrastructure/platform';
 
 const DEFAULT_MODEL_BASE_URL = 'https://huggingface.co/Bei0001/PP-DocLayoutV3-ONNX/resolve/main';
 const PDF_LAYOUT_MODEL_BASE_URL_ENV = 'PDF_LAYOUT_MODEL_BASE_URL';

compute-worker/src/inference/pdf/normalize-text.ts

@@ -1,5 +1,5 @@
 import type { TextItem } from 'pdfjs-dist/types/src/display/api';
-import type { PdfTextItem } from './types';
+import type { PdfTextItem } from './document-layout';
 
 interface ViewportLike {
   height: number;

compute-worker/src/inference/pdf/parse.ts

@@ -1,10 +1,10 @@
 import type { TextItem } from 'pdfjs-dist/types/src/display/api';
-import type { ParsedPdfDocument, ParsedPdfPage } from '../types/parsed-pdf';
+import type { ParsedPdfDocument, ParsedPdfPage } from '../../api/types';
 import { ensureModel } from './model';
-import { runLayoutModel } from './runLayoutModel';
+import { runLayoutModel } from './layout-model';
-import { mergeTextWithRegions } from './merge';
+import { mergeTextWithRegions } from './document-layout';
-import { configurePdfjsNodeRuntime, resolvePdfjsStandardFontDataUrl } from './pdfjs-runtime';
+import { configurePdfjsNodeRuntime, resolvePdfjsStandardFontDataUrl } from './pdfjs';
-import { PDF_PARSER_VERSION } from './parser-version';
+import { PDF_PARSER_VERSION } from '../../api/contracts';
 import { stitchCrossPageBlocks } from './stitch';
 import { renderPage } from './render';
 import { normalizeTextItemsForLayout } from './normalize-text';

compute-worker/src/inference/pdf/parser-version-key.ts

@@ -1,9 +0,0 @@
-import { PDF_PARSER_VERSION } from './parser-version';
-
-export function encodeParserVersion(
-  parserVersion: string,
-  defaultVersion = PDF_PARSER_VERSION,
-): string {
-  const normalized = parserVersion.trim() || defaultVersion;
-  return encodeURIComponent(normalized);
-}

compute-worker/src/inference/pdf/parser-version.ts

@@ -1 +0,0 @@
-export const PDF_PARSER_VERSION = 'pp-doclayoutv3-onnx@800+pdfjs@4.8.69';

compute-worker/src/inference/pdf/render.ts

@@ -1,5 +1,5 @@
 import type { Canvas } from '@napi-rs/canvas';
-import { configurePdfjsNodeRuntime, resolvePdfjsStandardFontDataUrl } from './pdfjs-runtime';
+import { configurePdfjsNodeRuntime, resolvePdfjsStandardFontDataUrl } from './pdfjs';
 
 type CanvasRuntime = {
   DOMMatrixCtor: unknown;

compute-worker/src/inference/pdf/stitch.ts

@@ -1,4 +1,4 @@
-import type { ParsedPdfDocument, ParsedPdfBlock } from '../types/parsed-pdf';
+import type { ParsedPdfDocument, ParsedPdfBlock } from '../../api/types';
 
 const STITCHABLE_KINDS = new Set<ParsedPdfBlock['kind']>([
   'text',

compute-worker/src/inference/pdf/types.ts

@@ -1,15 +0,0 @@
-import type { ParsedPdfBlockKind } from '../types/parsed-pdf';
-
-export interface PdfTextItem {
-  text: string;
-  x: number;
-  y: number;
-  width: number;
-  height: number;
-}
-
-export interface LayoutRegion {
-  bbox: [number, number, number, number];
-  label: ParsedPdfBlockKind;
-  confidence?: number;
-}

compute-worker/src/inference/platform/docstore.ts

@@ -1,21 +0,0 @@
-import fs from 'fs';
-import path from 'path';
-
-function findMonorepoRoot(startDir: string): string | null {
-  let current = path.resolve(startDir);
-  for (;;) {
-    const marker = path.join(current, 'pnpm-workspace.yaml');
-    if (fs.existsSync(marker)) return current;
-    const parent = path.dirname(current);
-    if (parent === current) return null;
-    current = parent;
-  }
-}
-
-function resolveDocstoreDir(): string {
-  const repoRoot = findMonorepoRoot(process.cwd());
-  if (repoRoot) return path.join(repoRoot, 'docstore');
-  return path.join(process.cwd(), 'docstore');
-}
-
-export const DOCSTORE_DIR = resolveDocstoreDir();

compute-worker/src/inference/types/index.ts

@@ -1,17 +0,0 @@
-export type {
-  TTSAudioBuffer,
-  TTSAudioBytes,
-  TTSSentenceAlignment,
-  TTSSentenceWord,
-} from './tts';
-
-export type {
-  ParsedPdfBlock,
-  ParsedPdfBlockFragment,
-  ParsedPdfBlockKind,
-  ParsedPdfDocument,
-  ParsedPdfPage,
-  PdfParsePhase,
-  PdfParseProgress,
-  PdfParseStatus,
-} from './parsed-pdf';

compute-worker/src/inference/types/tts.ts

@@ -1,16 +0,0 @@
-export type TTSAudioBuffer = ArrayBuffer;
-export type TTSAudioBytes = number[];
-
-export interface TTSSentenceWord {
-  text: string;
-  startSec: number;
-  endSec: number;
-  charStart: number;
-  charEnd: number;
-}
-
-export interface TTSSentenceAlignment {
-  sentence: string;
-  sentenceIndex: number;
-  words: TTSSentenceWord[];
-}

compute-worker/src/inference/whisper/align.ts

@@ -7,19 +7,18 @@ import { fileURLToPath } from 'url';
 import * as ort from 'onnxruntime-node';
 import { Tokenizer } from '@huggingface/tokenizers';
 import JSZip from 'jszip';
-import type { TTSAudioBuffer, TTSAudioBytes, TTSSentenceAlignment } from '../types/tts';
+import type { TTSAudioBuffer, TTSAudioBytes, TTSSentenceAlignment } from '../../api/types';
-import { getFFmpegPath } from '../platform/ffmpeg';
+import { getFFmpegPath } from '../../infrastructure/platform';
-import { getOnnxThreadsPerJob } from '../config/cpu-budget';
+import { getOnnxThreadsPerJob } from '../../infrastructure/config';
-import { getComputeTimeoutConfig } from '../config/timeout';
+import { getComputeTimeoutConfig } from '../../infrastructure/config';
 import {
   mapWordsToSentenceOffsets,
   type WhisperWord,
-} from './alignment-map';
+} from './timestamps';
-import { buildGoertzelCoefficients, goertzelPower } from './spectral';
 import {
   buildWordsFromTimestampedTokens,
   extractTokenStartTimestamps,
-} from './token-timestamps';
+} from './timestamps';
 import {
   ensureWhisperModel,
   WHISPER_CONFIG_PATH,
@@ -31,6 +30,26 @@ import {
   WHISPER_DECODER_WITH_PAST_MODEL_PATH,
 } from './model';
 
+export function buildGoertzelCoefficients(freqBins: number, fftSize: number): Float64Array {
+  const coeffs = new Float64Array(freqBins);
+  for (let k = 0; k < freqBins; k += 1) {
+    coeffs[k] = 2 * Math.cos((2 * Math.PI * k) / fftSize);
+  }
+  return coeffs;
+}
+
+export function goertzelPower(samples: Float32Array, coeff: number): number {
+  let s1 = 0;
+  let s2 = 0;
+  for (let i = 0; i < samples.length; i += 1) {
+    const s0 = samples[i] + (coeff * s1) - s2;
+    s2 = s1;
+    s1 = s0;
+  }
+  const power = (s1 * s1) + (s2 * s2) - (coeff * s1 * s2);
+  return !Number.isFinite(power) || power < 0 ? 0 : power;
+}
+
 interface WhisperAlignmentOptions {
   lang?: string;
   textHint?: string;

compute-worker/src/inference/whisper/alignment-map.ts

@@ -1,44 +0,0 @@
-import type { TTSSentenceAlignment, TTSSentenceWord } from '../types/tts';
-
-export interface WhisperWord {
-  start: number;
-  end: number;
-  word: string;
-}
-
-export function mapWordsToSentenceOffsets(sentence: string, words: WhisperWord[]): TTSSentenceAlignment {
-  const lowerSentence = sentence.toLowerCase();
-  let cursor = 0;
-
-  const alignedWords: TTSSentenceWord[] = words.map((w) => {
-    const token = w.word.trim();
-    if (!token) {
-      return {
-        text: '',
-        startSec: w.start,
-        endSec: w.end,
-        charStart: cursor,
-        charEnd: cursor,
-      };
-    }
-
-    const idx = lowerSentence.indexOf(token.toLowerCase(), cursor);
-    const start = idx >= 0 ? idx : cursor;
-    const end = Math.min(sentence.length, start + token.length);
-    cursor = Math.max(cursor, end);
-
-    return {
-      text: token,
-      startSec: w.start,
-      endSec: w.end,
-      charStart: start,
-      charEnd: end,
-    };
-  }).filter((word) => word.text.length > 0);
-
-  return {
-    sentence,
-    sentenceIndex: 0,
-    words: alignedWords,
-  };
-}

compute-worker/src/inference/whisper/model.ts

@@ -3,7 +3,7 @@ import { fileURLToPath } from 'url';
 import { createHash } from 'crypto';
 import { readFileSync } from 'fs';
 import { access, copyFile, mkdir, readFile, rename, unlink, writeFile } from 'fs/promises';
-import { DOCSTORE_DIR } from '../platform/docstore';
+import { DOCSTORE_DIR } from '../../infrastructure/platform';
 
 const MODULE_DIR = path.dirname(fileURLToPath(import.meta.url));
 const MODEL_DIR = path.join(DOCSTORE_DIR, 'model', 'whisper-base_timestamped');

compute-worker/src/inference/whisper/spectral.ts

@@ -1,21 +0,0 @@
-export function buildGoertzelCoefficients(freqBins: number, fftSize: number): Float64Array {
-  const coeffs = new Float64Array(freqBins);
-  for (let k = 0; k < freqBins; k += 1) {
-    coeffs[k] = 2 * Math.cos((2 * Math.PI * k) / fftSize);
-  }
-  return coeffs;
-}
-
-export function goertzelPower(samples: Float32Array, coeff: number): number {
-  let s1 = 0;
-  let s2 = 0;
-  for (let i = 0; i < samples.length; i += 1) {
-    const s0 = samples[i] + (coeff * s1) - s2;
-    s2 = s1;
-    s1 = s0;
-  }
-
-  const power = (s1 * s1) + (s2 * s2) - (coeff * s1 * s2);
-  if (!Number.isFinite(power) || power < 0) return 0;
-  return power;
-}

compute-worker/src/inference/whisper/timestamps.ts

@@ -1,5 +1,6 @@
 import type { Tokenizer } from '@huggingface/tokenizers';
 import type * as ort from 'onnxruntime-node';
+import type { TTSSentenceAlignment, TTSSentenceWord } from '../../api/types';
 
 const PUNCTUATION_REGEX = '\\p{P}\\u0021-\\u002F\\u003A-\\u0040\\u005B-\\u0060\\u007B-\\u007E';
 const PUNCTUATION_ONLY_REGEX = new RegExp(`^[${PUNCTUATION_REGEX}]+$`, 'gu');
@@ -12,6 +13,29 @@ export interface WhisperWordTiming {
   endSec: number;
 }
 
+export interface WhisperWord {
+  start: number;
+  end: number;
+  word: string;
+}
+
+export function mapWordsToSentenceOffsets(sentence: string, words: WhisperWord[]): TTSSentenceAlignment {
+  const lowerSentence = sentence.toLowerCase();
+  let cursor = 0;
+  const alignedWords: TTSSentenceWord[] = words.map((word) => {
+    const token = word.word.trim();
+    if (!token) {
+      return { text: '', startSec: word.start, endSec: word.end, charStart: cursor, charEnd: cursor };
+    }
+    const index = lowerSentence.indexOf(token.toLowerCase(), cursor);
+    const start = index >= 0 ? index : cursor;
+    const end = Math.min(sentence.length, start + token.length);
+    cursor = Math.max(cursor, end);
+    return { text: token, startSec: word.start, endSec: word.end, charStart: start, charEnd: end };
+  }).filter((word) => word.text.length > 0);
+  return { sentence, sentenceIndex: 0, words: alignedWords };
+}
+
 function medianFilter(data: Float32Array, windowSize: number): Float32Array {
   if (windowSize % 2 === 0 || windowSize <= 0) {
     throw new Error('Window size must be a positive odd number');

compute-worker/src/infrastructure/config.ts

@@ -117,3 +117,23 @@ export async function withIdleTimeoutAndHardCap<T>(input: IdleTimeoutAndHardCapI
     throw error;
   }
 }
+
+export function getComputeJobConcurrency(): number {
+  return readPositiveIntEnv('COMPUTE_JOB_CONCURRENCY', 1);
+}
+
+export function getAvailableCpuCores(): number {
+  if (typeof os.availableParallelism === 'function') {
+    const value = os.availableParallelism();
+    if (Number.isFinite(value) && value >= 1) return Math.floor(value);
+  }
+  const fallback = os.cpus().length;
+  return Number.isFinite(fallback) && fallback >= 1 ? Math.floor(fallback) : 1;
+}
+
+export function getOnnxThreadsPerJob(): number {
+  const concurrency = getComputeJobConcurrency();
+  const usableCores = Math.max(1, getAvailableCpuCores() - 1);
+  return Math.max(1, Math.floor(usableCores / concurrency));
+}
+import os from 'node:os';

compute-worker/src/infrastructure/platform.ts

@@ -1,6 +1,26 @@
-import { existsSync } from 'fs';
+import fs from 'fs';
+import path from 'path';
 import ffmpegStatic from 'ffmpeg-static';
 
+function findMonorepoRoot(startDir: string): string | null {
+  let current = path.resolve(startDir);
+  for (;;) {
+    const marker = path.join(current, 'pnpm-workspace.yaml');
+    if (fs.existsSync(marker)) return current;
+    const parent = path.dirname(current);
+    if (parent === current) return null;
+    current = parent;
+  }
+}
+
+function resolveDocstoreDir(): string {
+  const repoRoot = findMonorepoRoot(process.cwd());
+  if (repoRoot) return path.join(repoRoot, 'docstore');
+  return path.join(process.cwd(), 'docstore');
+}
+
+export const DOCSTORE_DIR = resolveDocstoreDir();
+
 function normalizePath(value: unknown): string | null {
   if (typeof value !== 'string') return null;
   const trimmed = value.trim();
@@ -9,20 +29,17 @@ function normalizePath(value: unknown): string | null {
 
 function resolveBinary(envValue: string | null, bundledValue: string | null, envVarName: string, packageName: string): string {
   if (envValue) {
-    if ((envValue.includes('/') || envValue.includes('\\')) && !existsSync(envValue)) {
+    if ((envValue.includes('/') || envValue.includes('\\')) && !fs.existsSync(envValue)) {
       throw new Error(`${envVarName} points to a missing binary: ${envValue}`);
     }
     return envValue;
   }
-
   if (!bundledValue) {
     throw new Error(`${packageName} binary is unavailable on this platform. Set ${envVarName} to an installed binary path.`);
   }
-
+  if ((bundledValue.includes('/') || bundledValue.includes('\\')) && !fs.existsSync(bundledValue)) {
-  if ((bundledValue.includes('/') || bundledValue.includes('\\')) && !existsSync(bundledValue)) {
     throw new Error(`${packageName} resolved to a missing binary path: ${bundledValue}`);
   }
-
   return bundledValue;
 }
 

compute-worker/src/jobs/handlers.ts

@@ -2,8 +2,8 @@ import { z } from 'zod';
 import {
   runPdfLayoutFromPdfBuffer,
   runWhisperAlignmentFromAudioBuffer,
-} from '../inference/local-runtime';
+} from '../inference/runtime';
-import { withIdleTimeoutAndHardCap, withTimeout } from '../inference';
+import { withIdleTimeoutAndHardCap, withTimeout } from '../infrastructure/config';
 import type {
   PdfLayoutJobRequest,
   PdfLayoutJobResult,

compute-worker/src/storage/artifact-addressing.ts

@@ -1,5 +1,5 @@
-import { PDF_PARSER_VERSION } from '../inference/pdf/parser-version';
+import { PDF_PARSER_VERSION } from '../api/contracts';
-import { encodeParserVersion } from '../inference/pdf/parser-version-key';
+import { encodeParserVersion } from '../api/contracts';
 
 const DOCUMENT_ID_REGEX = /^[a-f0-9]{64}$/i;
 const SAFE_NAMESPACE_REGEX = /^[a-zA-Z0-9._-]{1,128}$/;

compute-worker/tests/compute/algorithms/pdf-merge-text-with-regions.test.ts

@@ -1,5 +1,5 @@
 import { describe, expect, test } from 'vitest';
-import { mergeTextWithRegions } from '../../../src/inference';
+import { mergeTextWithRegions } from '../../../src/inference/pdf/document-layout';
 
 describe('mergeTextWithRegions', () => {
   test('assigns text items to containing regions by centroid and joins in reading order', () => {

compute-worker/tests/compute/algorithms/pdf-parse-normalize-text-items.test.ts

@@ -1,6 +1,6 @@
 import { describe, expect, test } from 'vitest';
 
-import { normalizeTextItemsForLayout } from '../../../src/inference';
+import { normalizeTextItemsForLayout } from '../../../src/inference/pdf/normalize-text';
 import type { TextItem } from 'pdfjs-dist/types/src/display/api';
 
 function makeTextItem(

compute-worker/tests/compute/algorithms/pdf-stitch-cross-page-blocks.test.ts

@@ -1,6 +1,6 @@
 import { describe, expect, test } from 'vitest';
-import { stitchCrossPageBlocks } from '../../../src/inference';
+import { stitchCrossPageBlocks } from '../../../src/inference/pdf/stitch';
-import type { ParsedPdfBlock, ParsedPdfDocument, ParsedPdfBlockKind } from '../../../src/inference/types';
+import type { ParsedPdfBlock, ParsedPdfDocument, ParsedPdfBlockKind } from '../../../src/api/types';
 import { makeParsedPdfDocument, makeParsedPdfPage } from './support/document-fixtures';
 
 function makeBlock(

compute-worker/tests/compute/algorithms/support/document-fixtures.ts

@@ -1,5 +1,5 @@
 import type { BaseDocument } from '../../../../../src/types/documents';
-import type { ParsedPdfBlock, ParsedPdfBlockKind, ParsedPdfDocument, ParsedPdfPage } from '../../../../src/inference/types';
+import type { ParsedPdfBlock, ParsedPdfBlockKind, ParsedPdfDocument, ParsedPdfPage } from '../../../../src/api/types';
 
 export function makeBaseDocument(overrides: Partial<BaseDocument> = {}): BaseDocument {
   return {

compute-worker/tests/compute/algorithms/whisper-alignment-mapping.test.ts

@@ -1,7 +1,7 @@
 import { describe, expect, test } from 'vitest';
 import {
   mapWordsToSentenceOffsets,
-} from '../../../src/inference';
+} from '../../../src/inference/whisper/timestamps';
 
 describe('whisper alignment mapping', () => {
   test('maps words to sentence offsets with punctuation and repeated spaces', () => {

compute-worker/tests/compute/algorithms/whisper-spectral.test.ts

@@ -1,5 +1,5 @@
 import { describe, expect, test } from 'vitest';
-import { buildGoertzelCoefficients, goertzelPower } from '../../../src/inference';
+import { buildGoertzelCoefficients, goertzelPower } from '../../../src/inference/whisper/align';
 
 function dftPower(samples: Float32Array, k: number): number {
   const n = samples.length;

compute-worker/tests/compute/algorithms/whisper-token-timestamps.test.ts

@@ -3,7 +3,7 @@ import * as ort from 'onnxruntime-node';
 import {
   buildWordsFromTimestampedTokens,
   extractTokenStartTimestamps,
-} from '../../../src/inference';
+} from '../../../src/inference/whisper/timestamps';
 
 describe('whisper token timestamp alignment', () => {
   test('extracts monotonic token timestamps from cross-attention maps', () => {

compute-worker/tests/compute/control-plane/run-layout-model.test.ts

@@ -76,7 +76,7 @@ vi.mock('../../../src/inference/pdf/model', () => ({
   MODEL_PREPROCESSOR_PATH: '/tmp/model-preprocessor.json',
 }));
 
-vi.mock('../../../src/inference/config/cpu-budget', () => ({
+vi.mock('../../../src/infrastructure/config', () => ({
   getOnnxThreadsPerJob: vi.fn(() => 1),
 }));
 
@@ -107,7 +107,7 @@ describe('runLayoutModel', () => {
       },
     };
 
-    const { runLayoutModel } = await import('../../../src/inference/pdf/runLayoutModel');
+    const { runLayoutModel } = await import('../../../src/inference/pdf/layout-model');
     const regions = await runLayoutModel({
       pageWidth: 100,
       pageHeight: 100,
@@ -154,7 +154,7 @@ describe('runLayoutModel', () => {
       },
     };
 
-    const { runLayoutModel } = await import('../../../src/inference/pdf/runLayoutModel');
+    const { runLayoutModel } = await import('../../../src/inference/pdf/layout-model');
     const regions = await runLayoutModel({
       pageWidth: 100,
       pageHeight: 100,