@dxos/plugin-transcription 0.8.2-main.fbd8ed0 → 0.8.2-staging.7ac8446

package/src/components/TranscriptionSettings/index.ts

@@ -1,5 +0,0 @@
-//
-// Copyright 2025 DXOS.org
-//
-
-export * from './TranscriptionSettings';

package/src/entity-extraction/entity-extraction.test.ts

@@ -1,97 +0,0 @@
-//
-// Copyright 2025 DXOS.org
-//
-
-import { describe, test, expect } from 'vitest';
-
-import { AIServiceEdgeClient } from '@dxos/assistant';
-import { AI_SERVICE_ENDPOINT } from '@dxos/assistant/testing';
-import { log } from '@dxos/log';
-import { createTestData } from '@dxos/schema/testing';
-import { range } from '@dxos/util';
-
-import { processTranscriptMessage, postprocessText } from './entity-extraction';
-
-const aiService = new AIServiceEdgeClient({
-  endpoint: AI_SERVICE_ENDPOINT.REMOTE,
-});
-
-describe.skip('EntityExtraction', { timeout: 180_000 }, () => {
-  test.skip('should process a transcript block', async () => {
-    const { transcriptMessages, documents, contacts } = await createTestData();
-
-    log.info('context', { documents, contacts });
-
-    for (const message of transcriptMessages) {
-      log.info('input', message);
-      const { message: enhancedMessage } = await processTranscriptMessage({
-        message,
-        aiService,
-        context: {
-          objects: [...documents, ...Object.values(contacts)],
-        },
-      });
-      log.info('output', enhancedMessage);
-    }
-  });
-
-  test.only('computational irreducibility', async () => {
-    const { transcriptWoflram, documents, contacts } = await createTestData();
-
-    log.info('context', { documents, contacts });
-    const message = transcriptWoflram[0];
-    log.info('input', message);
-
-    await Promise.all(
-      range(10).map(async () => {
-        const { message: enhancedMessage, timeElapsed } = await processTranscriptMessage({
-          message,
-          aiService,
-          context: {
-            objects: [...documents, ...Object.values(contacts)],
-          },
-        });
-        log.info('output', { message: enhancedMessage.blocks[0], timeElapsed });
-      }),
-    );
-  });
-
-  test('org and document linking', async () => {
-    const { transcriptJosiah, documents, contacts, organizations } = await createTestData();
-
-    log.info('context', { contacts, organizations, documents });
-
-    for (const message of transcriptJosiah) {
-      log.info('input', message);
-
-      const { message: enhancedMessage } = await processTranscriptMessage({
-        message,
-        aiService,
-        context: {
-          objects: [...documents, ...Object.values(contacts), ...Object.values(organizations)],
-        },
-      });
-      log.info('output', enhancedMessage);
-    }
-  });
-});
-
-describe('postprocessText', () => {
-  test('should replace quotes with DXN references', () => {
-    const quotes = {
-      references: [{ quote: 'computational irreducibility', id: '01JTG9JW11XGWJZ32AW8ET93D1' }],
-    };
-
-    expect(postprocessText('This is a computational irreducibility test.', quotes)).toBe(
-      'This is a [computational irreducibility][dxn:echo:@:01JTG9JW11XGWJZ32AW8ET93D1] test.',
-    );
-    expect(
-      postprocessText(
-        "And what I'd like to talk today about is Steven Wolfram's concept of a computational irreducibility.",
-        quotes,
-      ),
-    ).toBe(
-      "And what I'd like to talk today about is Steven Wolfram's concept of a [computational irreducibility][dxn:echo:@:01JTG9JW11XGWJZ32AW8ET93D1].",
-    );
-  });
-});

package/src/entity-extraction/entity-extraction.ts

@@ -1,149 +0,0 @@
-//
-// Copyright 2025 DXOS.org
-//
-
-import { Schema } from 'effect';
-
-import { createTemplate, Message, structuredOutputParser } from '@dxos/artifact';
-import type { AIServiceClient } from '@dxos/assistant';
-import { MixedStreamParser } from '@dxos/assistant';
-import { asyncTimeout } from '@dxos/async';
-import { type BaseEchoObject, create, ObjectId } from '@dxos/echo-schema';
-import { DXN } from '@dxos/keys';
-import { log } from '@dxos/log';
-import { DataType } from '@dxos/schema';
-
-import SYSTEM_PROMPT from './system-prompt.tpl?raw';
-
-type ProcessTranscriptMessageParams = {
-  message: DataType.Message;
-  aiService: AIServiceClient;
-  context: {
-    objects?: BaseEchoObject[];
-  };
-
-  options?: {
-    /**
-     * Timeout for the entity extraction.
-     */
-    timeout?: number;
-
-    /**
-     * Fallback to raw text if the entity extraction fails.
-     * Otherwise the function will throw an error.
-     * @default false
-     */
-    fallbackToRaw?: boolean;
-  };
-};
-
-type ProcessTranscriptMessageResult = {
-  message: DataType.Message;
-  timeElapsed: number;
-};
-
-/**
- * Process Handlebars template.
- */
-const createSystemPrompt = (): string => {
-  const template = createTemplate(SYSTEM_PROMPT);
-  return template({});
-};
-
-const ReferencedQuotes = Schema.Struct({
-  references: Schema.Array(
-    Schema.Struct({
-      quote: Schema.String,
-      id: Schema.String,
-    }),
-  ).annotations({
-    description:
-      'The references to the context objects that are mentioned in the transcript. quote should match the original transcript text exactly, while id is the id of the context object.',
-  }),
-});
-interface ReferencedQuotes extends Schema.Schema.Type<typeof ReferencedQuotes> {}
-
-// TODO(dmaretskyi): Move context to a vector search index.
-export const processTranscriptMessage = async (
-  params: ProcessTranscriptMessageParams,
-): Promise<ProcessTranscriptMessageResult> => {
-  try {
-    const outputParser = structuredOutputParser(ReferencedQuotes);
-    const runParser = async (): Promise<ProcessTranscriptMessageResult> => {
-      const startTime = performance.now();
-      const result = outputParser.getResult(
-        await new MixedStreamParser().parse(
-          await params.aiService.execStream({
-            model: '@anthropic/claude-3-5-haiku-20241022',
-            systemPrompt: createSystemPrompt(),
-            history: [
-              create(Message, {
-                role: 'user',
-                content: [
-                  {
-                    type: 'text',
-                    text: `<context>${JSON.stringify(params.context.objects)}</context>`,
-                  },
-                  {
-                    type: 'text',
-                    text: `<transcript>${JSON.stringify(params.message.blocks)}</transcript>`,
-                  },
-                ],
-              }),
-            ],
-            tools: [outputParser.tool],
-          }),
-        ),
-      );
-      log.info('entity extraction result', { refs: result.references });
-
-      return {
-        message: create(DataType.Message, {
-          ...params.message,
-          blocks: params.message.blocks.map((block, i) =>
-            block.type !== 'transcription'
-              ? block
-              : {
-                  ...block,
-                  text: postprocessText(block.text, result),
-                },
-          ),
-        }),
-        timeElapsed: performance.now() - startTime,
-      };
-    };
-
-    if (params.options?.timeout && params.options.timeout > 0) {
-      return await asyncTimeout(runParser(), params.options.timeout);
-    } else {
-      return await runParser();
-    }
-  } catch (error) {
-    if (params.options?.fallbackToRaw) {
-      log.warn('failed to process transcript message, falling back to raw text', { error });
-      return {
-        message: params.message,
-        timeElapsed: 0,
-      };
-    } else {
-      throw error;
-    }
-  }
-};
-
-/**
- * Finds and replaces all quotes with DXNs references.
- */
-// TODO(dmaretskyi): Lookup and verifiy ids from provided context.
-export const postprocessText = (text: string, quotes: ReferencedQuotes) => {
-  for (const quote of quotes.references) {
-    if (!ObjectId.isValid(quote.id)) {
-      continue;
-    }
-
-    // Use a case-insensitive regular expression to replace the quote
-    const regex = new RegExp(quote.quote.replace(/[.*+?^${}()|[\]\\]/g, '\\$&'), 'gi');
-    text = text.replace(regex, `[${quote.quote}][${DXN.fromLocalObjectId(quote.id).toString()}]`);
-  }
-  return text;
-};

package/src/entity-extraction/index.ts

@@ -1,5 +0,0 @@
-//
-// Copyright 2025 DXOS.org
-//
-
-export * from './entity-extraction';

package/src/entity-extraction/system-prompt.tpl

@@ -1,38 +0,0 @@
-You are an AI assistant specialized in processing and enhancing transcripts with contextual information.
-Your task is to analyze a given transcript and extract references to the context objects.
-
-The context and transcript are provided in the user message in `context` and `transcript` XML tags.
-
-Here are your instructions:
-
-1. Read the transcript and context information provided above.
-
-2. Process each segment of the transcript separately.
-
-3. For each segment:
-   a. Analyze the content and identify relevant information from the context.
-   b. Extract a number of references to the context objects that are mentioned in the segment.
-   c. Output the references in the format `{ "quote": "earnings report", "id": "01JT0JP9AX0XKGZX4MV4B69VT6" }`.
-   d. If there are no references, just output an empty array.
-
-4. Maintain the exact order and structure of the original transcript segments.
-
-Output Format:
-The final output should be a JSON array of references.
-
-Example output structure:
-```json
-{
-  "references": [
-    { "quote": "earnings report", "id": "01JT0JP9AX0XKGZX4MV4B69VT6" },
-    { "quote": "John", "id": "01JT0VM0VN3ZKN7GDBM8CV9BA7" }
-  ]
-}
-```
-
-Remember:
-- Only quote the text that is in the transcript.
-- Analyze each segment carefully before processing it.
-- Only reference objects that appear in the context and exactly by their id, If the object is not in the context, do not insert a reference. 
-
-Now, process the transcript and context provided below according to these instructions.

package/src/entity-extraction/types.d.ts

@@ -1,8 +0,0 @@
-//
-// Copyright 2025 DXOS.org
-//
-
-declare module '*.tpl?raw' {
-  const content: string;
-  export default content;
-}

package/src/hooks/useQueueModelAdapter.ts

@@ -1,63 +0,0 @@
-//
-// Copyright 2025 DXOS.org
-//
-
-import { useMemo, useEffect, useState } from 'react';
-
-import { type Queue } from '@dxos/react-client/echo';
-
-import { type Chunk, type ChunkRenderer, SerializationModel } from '../model';
-
-/**
- * Model adapter for a queue.
- */
-export const useQueueModelAdapter = <T extends Chunk>(
-  renderer: ChunkRenderer<T>,
-  queue: Queue<T> | undefined,
-  initialChunks: T[] = [],
-): SerializationModel<T> => {
-  const model = useMemo(() => new SerializationModel<T>(renderer, initialChunks), [queue]);
-  const [loaded, setLoaded] = useState(false);
-
-  // Set initial blocks.
-  useEffect(() => {
-    if (!queue) {
-      return;
-    }
-
-    const update = () => {
-      for (const block of queue?.items ?? []) {
-        model.appendChunk(block);
-      }
-
-      setLoaded(true);
-    };
-
-    // TODO(burdon): This is a hack to ensure the queue is loaded.
-    let i: NodeJS.Timeout | undefined;
-    if (queue.isLoading) {
-      i = setInterval(() => {
-        if (!queue?.isLoading) {
-          clearInterval(i);
-          update();
-        }
-      }, 1_000);
-    } else {
-      update();
-    }
-
-    return () => clearTimeout(i);
-  }, [model, queue]);
-
-  // TODO(burdon): Can we listen for queue events?
-  useEffect(() => {
-    if (!loaded || !queue || model.chunks.length === queue.items.length) {
-      return;
-    }
-
-    const chunk = queue.items[queue.items.length - 1];
-    model.appendChunk(chunk);
-  }, [model, loaded, queue?.items.length]);
-
-  return model;
-};

package/src/model/index.ts

@@ -1,5 +0,0 @@
-//
-// Copyright 2025 DXOS.org
-//
-
-export * from './model';

package/src/model/model.test.ts

@@ -1,169 +0,0 @@
-//
-// Copyright 2025 DXOS.org
-//
-
-import { EditorView } from '@codemirror/view';
-import { describe, test } from 'vitest';
-
-import { type DataType } from '@dxos/schema';
-
-import { SerializationModel, DocumentAdapter, type ChunkRenderer } from './model';
-
-const blockToMarkdown: ChunkRenderer<DataType.Message> = (
-  message: DataType.Message,
-  index: number,
-  debug = true,
-): string[] => {
-  return [
-    `###### ${message.sender.name}`,
-    ...message.blocks.filter((block) => block.type === 'transcription').map((block) => block.text),
-    '',
-  ];
-};
-
-const createDate = () => new Date().toISOString();
-
-describe('SerializationModel', () => {
-  test('basic', ({ expect }) => {
-    const model = new SerializationModel<DataType.Message>(blockToMarkdown);
-    expect(model.chunks.length).to.eq(0);
-    expect(model.doc.toString()).to.eq('');
-
-    // Create message.
-    const message: DataType.Message = {
-      id: '1',
-      created: createDate(),
-      sender: { name: 'Alice' },
-      blocks: [
-        {
-          type: 'transcription',
-          started: createDate(),
-          text: 'Hello world!',
-        },
-      ],
-    };
-    model.appendChunk(message);
-    {
-      const text = model.doc.toString();
-      expect(text).to.eq('###### Alice\nHello world!\n\n');
-      expect(model.chunks.length).to.eq(1);
-    }
-
-    // Update message.
-    message.blocks.push({ type: 'transcription', started: createDate(), text: 'Hello again!' });
-    model.updateChunk(message);
-    {
-      const text = model.doc.toString();
-      expect(text).to.eq('###### Alice\nHello world!\nHello again!\n\n');
-      expect(model.chunks.length).to.eq(1);
-    }
-  });
-
-  test('sync - append', ({ expect }) => {
-    const view = new EditorView({ extensions: [], doc: '' });
-    const model = new SerializationModel<DataType.Message>(blockToMarkdown);
-    const adapter = new DocumentAdapter(view);
-    expect(adapter.lineCount()).to.eq(0);
-    expect(view.state.doc.toString()).to.eq('');
-
-    // Append message.
-    {
-      const message: DataType.Message = {
-        id: '1',
-        created: createDate(),
-        sender: { name: 'Alice' },
-        blocks: [
-          {
-            type: 'transcription',
-            started: createDate(),
-            text: 'Hello world!',
-          },
-        ],
-      };
-      model.appendChunk(message);
-      model.sync(adapter);
-      expect(view.state.doc.toString()).to.eq('###### Alice\nHello world!\n\n');
-    }
-
-    // Append message.
-    {
-      const message: DataType.Message = {
-        id: '2',
-        created: createDate(),
-        sender: { name: 'Bob' },
-        blocks: [
-          {
-            type: 'transcription',
-            started: createDate(),
-            text: 'Hello world!',
-          },
-        ],
-      };
-      model.appendChunk(message);
-      model.sync(adapter);
-      expect(view.state.doc.toString()).to.eq('###### Alice\nHello world!\n\n###### Bob\nHello world!\n\n');
-    }
-  });
-
-  test('sync - append, update, delete', ({ expect }) => {
-    const view = new EditorView({ extensions: [], doc: '' });
-    const model = new SerializationModel<DataType.Message>(blockToMarkdown);
-    const adapter = new DocumentAdapter(view);
-    expect(adapter.lineCount()).to.eq(0);
-
-    // Append message.
-    {
-      const message: DataType.Message = {
-        id: '1',
-        created: createDate(),
-        sender: { name: 'Alice' },
-        blocks: [
-          {
-            type: 'transcription',
-            started: createDate(),
-            text: 'Hello world!',
-          },
-        ],
-      };
-      model.appendChunk(message);
-      model.sync(adapter);
-      expect(view.state.doc.toString()).to.eq('###### Alice\nHello world!\n\n');
-
-      // Update message (add block).
-      message.blocks.push({ type: 'transcription', started: createDate(), text: 'Hello again!' });
-      model.updateChunk(message);
-      model.sync(adapter);
-      expect(view.state.doc.toString()).to.eq('###### Alice\nHello world!\nHello again!\n\n');
-
-      // Update message (remove block).
-      message.blocks.splice(0, 1);
-      model.updateChunk(message);
-      model.sync(adapter);
-      expect(view.state.doc.toString()).to.eq('###### Alice\nHello again!\n\n');
-
-      expect(model.getChunkAtLine(1)).to.eq(message);
-    }
-
-    // Append message.
-    {
-      const message: DataType.Message = {
-        id: '2',
-        created: createDate(),
-        sender: { name: 'Bob' },
-        blocks: [{ type: 'transcription', started: createDate(), text: 'Hello again!' }],
-      };
-      model.appendChunk(message);
-      model.sync(adapter);
-      expect(view.state.doc.toString()).to.eq('###### Alice\nHello again!\n\n###### Bob\nHello again!\n\n');
-
-      expect(model.getChunkAtLine(4)).to.eq(message);
-    }
-
-    // Delete block.
-    {
-      model.deleteBlock('1');
-      model.sync(adapter);
-      expect(view.state.doc.toString()).to.eq('###### Bob\nHello again!\n\n');
-    }
-  });
-});