Transcript Analyzer

{
  "name": "transcript-analyzer",
  "description": "This skill analyzes meeting transcripts to extract decisions, action items, opinions, questions, and terminology using C",
  "author": {
    "name": "Gleb Kalinin"
  },
  "repository": "https://github.com/glebis/claude-skills",
  "license": "MIT"
}

#!/usr/bin/env node
import path from 'path';
import { promises as fs } from 'fs';
import { processTranscript } from '@/lib/extract-service';
import {
  buildFullDocument
} from '@/lib/markdown';
import { Extraction, GlossaryTerm } from '@/types';

interface CLIOptions {
  file?: string;
  glossaryPath?: string;
  skipGlossary?: boolean;
  includeTranscript?: boolean;
  noExtractions?: boolean;
  noGlossary?: boolean;
  output?: string;
  maxTerms?: number;
  chunkSize?: number;
}

async function main() {
  const options = parseArgs(process.argv.slice(2));

  if (!options.file) {
    printHelp();
    process.exit(1);
  }

  console.log('📄 Reading transcript...');
  const transcriptPath = path.resolve(options.file);
  const transcript = await fs.readFile(transcriptPath, 'utf-8');
  console.log(`   Loaded ${transcript.length} characters`);

  console.log('📚 Loading glossary...');
  const glossary = await loadGlossary(options);
  console.log(`   Found ${glossary.length} existing terms`);

  console.log('🤖 Processing transcript with AI...');
  const result = await processTranscript({
    transcript,
    glossary,
    chunkSize: options.chunkSize,
    maxTerms: options.maxTerms
  });

  console.log(`✓ Processing complete`);
  console.log(`   Chunks: ${result.stats.chunks}`);
  console.log(`   Extractions: ${result.stats.totalExtractions}`);
  console.log(`   New terms: ${result.stats.newTerms}`);
  if (result.stats.inputTokens && result.stats.outputTokens) {
    console.log(`   Tokens: ${result.stats.inputTokens + result.stats.outputTokens} total`);
  }

  console.log('📝 Building markdown document...');
  const markdown = buildFullDocument({
    transcript,
    extractions: result.extractions,
    glossaryTerms: glossary,
    suggestedTerms: result.suggestedTerms,
    stats: result.stats,
    includeTranscript: options.includeTranscript,
    includeExtractions: !options.noExtractions,
    includeGlossary: !options.noGlossary
  });

  if (options.output) {
    const outputPath = path.resolve(options.output);
    await fs.writeFile(outputPath, markdown, 'utf-8');
    console.log(`✓ Exported to ${outputPath}`);
  } else {
    console.log(markdown);
  }
}

function parseArgs(argv: string[]): CLIOptions {
  const options: CLIOptions = {};
  const args = [...argv];

  while (args.length > 0) {
    const arg = args.shift()!;

    if (!arg.startsWith('-') && !options.file) {
      options.file = arg;
      continue;
    }

    switch (arg) {
      case '--file':
      case '-f':
        options.file = args.shift();
        break;
      case '--glossary':
        options.glossaryPath = args.shift();
        break;
      case '--skip-glossary':
        options.skipGlossary = true;
        break;
      case '--include-transcript':
        options.includeTranscript = true;
        break;
      case '--no-extractions':
        options.noExtractions = true;
        break;
      case '--no-glossary':
        options.noGlossary = true;
        break;
      case '--output':
      case '-o':
        options.output = args.shift();
        break;
      case '--max-terms':
        options.maxTerms = Number(args.shift());
        break;
      case '--chunk-size':
        options.chunkSize = Number(args.shift());
        break;
      case '--help':
      case '-h':
        printHelp();
        process.exit(0);
      default:
        if (!options.file) {
          options.file = arg;
        }
        break;
    }
  }

  return options;
}

async function loadGlossary(options: CLIOptions): Promise<GlossaryTerm[]> {
  if (options.skipGlossary) return [];

  const defaultPath = path.join(process.cwd(), 'data', 'glossary.json');
  const target = options.glossaryPath
    ? path.resolve(options.glossaryPath)
    : defaultPath;

  try {
    const data = await fs.readFile(target, 'utf-8');
    return JSON.parse(data);
  } catch {
    return [];
  }
}

function printHelp() {
  console.log(`Usage: npm run cli -- <transcript.txt> -o <output.md> [options]

Outputs a Markdown document with metadata, extractions, and glossary.

Options:
  <file>               Transcript file to analyze (first positional arg)
  -o, --output <path>  Write markdown to a file instead of stdout
  --glossary <path>    Path to glossary JSON (defaults to data/glossary.json)
  --skip-glossary      Do not preload glossary terms
  --include-transcript Include full transcript in output [default: off]
  --no-extractions     Exclude extractions section from output
  --no-glossary        Exclude glossary section from output
  --max-terms <num>    Maximum glossary suggestions
  --chunk-size <num>   Override chunk size for processing
  -h, --help           Show this help message

Examples:
  npm run cli -- transcript.txt -o output.md
  npm run cli -- transcript.txt -o output.md --include-transcript
  npm run cli -- transcript.txt --no-glossary -o output.md

Output includes:
  - YAML frontmatter with processing metadata (chunks, tokens, model)
  - Full transcript [default: off, use --include-transcript]
  - Extractions (decisions, actions, opinions, questions, terms) [default: on]
  - Glossary terms (approved + suggested) [default: on]
`);
}

main().catch((error) => {
  console.error(error);
  process.exit(1);
});

[
  {
    "id": "term-1764268209078",
    "term": "Oldowan tradition",
    "definition": "The earliest known stone tool technology, dating from 2.6+ million years ago, characterized by simple stone tools that required planning and foresight to manufacture",
    "aliases": [
      "Oldowan tools"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268210888",
    "term": "Acheulean tradition",
    "definition": "A stone tool technology spanning 1.7-0.1 million years ago, characterized by symmetrical hand axes that demanded greater motor control and cognitive sophistication than earlier tools",
    "aliases": [
      "Acheulean hand axes"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268212895",
    "term": "embodied cognition",
    "definition": "The concept that thinking occurs through physical interaction with the world, particularly relevant to tool-making where cognitive development happens through hands-on manipulation and creation",
    "aliases": [
      "embodied knowledge"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268214559",
    "term": "manual dexterity",
    "definition": "Fine motor control and hand-brain coordination that was both required for and developed by tool-making, creating evolutionary pressure for increasingly sophisticated motor capabilities",
    "aliases": [
      "fine motor control"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268218126",
    "term": "executive function",
    "definition": "Higher-level cognitive processes including planning, foresight, and problem-solving that were enhanced through tool-making activities and drove brain development",
    "aliases": [],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268256469",
    "term": "Homo faber",
    "definition": "Hannah Arendt's philosophical concept defining humans by their fundamental ability to shape their environment and fate through making tools and objects, positioning tool-making as core human identity rather than mere capability",
    "aliases": [],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268257592",
    "term": "cognitive-physical feedback loop",
    "definition": "The evolutionary cycle where tool-making drove brain development through enhanced executive function and motor skills, which in turn enabled more sophisticated tool creation, creating selective pressure for larger brains",
    "aliases": [],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268683191",
    "term": "MCP",
    "definition": "Model Context Protocol - a standard for connecting AI models to external data sources and services, allowing models to access content from various applications and databases",
    "aliases": [
      "MCP-серверы",
      "MCP-сервер",
      "MCP-шки"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268684581",
    "term": "скиллы",
    "definition": "Модульные возможности, расширяющие функциональность Claude - папки с файлами, содержащие промпты и код для выполнения специфических задач",
    "aliases": [
      "agent skills",
      "Claude skills",
      "скилл"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268685606",
    "term": "Клод",
    "definition": "AI-ассистент от Anthropic, используемый для программирования и других задач",
    "aliases": [
      "Claude",
      "Клод Код",
      "Claude Code"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268686333",
    "term": "контекст",
    "definition": "Общая память сессии с моделью, в которую попадает история взаимодействий и которая влияет на производительность",
    "aliases": [
      "общий контекст"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268687364",
    "term": "скилл",
    "definition": "A self-contained package containing all actions and materials needed to teach Claude to perform a specific task, similar to a skill for a person. Can include text files, scripts, templates, and other resources.",
    "aliases": [
      "skill"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268689054",
    "term": "software 3.0",
    "definition": "AI-driven programming approach where interactions are managed through prompts and AI agents rather than traditional code",
    "aliases": [],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268690067",
    "term": "Obsidian",
    "definition": "A note-taking application that stores files locally, which can be integrated with AI assistants through MCP",
    "aliases": [],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268690993",
    "term": "Firecrawl",
    "definition": "A tool or service used for web scraping and content extraction, mentioned in context of a skill that can generate scientific templates",
    "aliases": [],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268692040",
    "term": "LLM",
    "definition": "Large Language Model - AI systems that can process and generate text, mentioned as an alternative to Claude for certain tasks",
    "aliases": [],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268695398",
    "term": "Claude",
    "definition": "An AI assistant that can write code and scripts, as referenced in the context of writing programming scripts",
    "aliases": [
      "him"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268697671",
    "term": "skill (скилл)",
    "definition": "A coding automation feature in Claude that can execute complex multi-step tasks, often converting from software 1.0 utilities and scripts to software 3.0 AI-driven tools",
    "aliases": [
      "skills",
      "скиллы"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764268705680",
    "term": "Claude Code",
    "definition": "A command-line interface tool for interacting with Claude AI, distinct from Claude Desktop which is a GUI application",
    "aliases": [
      "клод-код",
      "CloudCode",
      "Клод Код"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764269937176",
    "term": "стартап",
    "definition": "A startup company or early-stage business venture, often technology-focused",
    "aliases": [
      "стратап"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764269939004",
    "term": "предпринимательство",
    "definition": "Entrepreneurship - the activity of starting and running business ventures",
    "aliases": [],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764269940881",
    "term": "ценности ориентированных предпринимателей",
    "definition": "Values-oriented entrepreneurs who focus on meaningful impact rather than just profit",
    "aliases": [
      "про человека и сценностные истории"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764269944073",
    "term": "стартап",
    "definition": "A new business venture or entrepreneurial project that requires focus on money and revenue generation to succeed",
    "aliases": [
      "startup"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764269945756",
    "term": "self-concordance",
    "definition": "A psychological term referring to the degree of alignment between one's actions and personal values",
    "aliases": [],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764269948304",
    "term": "self-perception",
    "definition": "One's own perception or understanding of themselves and their characteristics",
    "aliases": [],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  },
  {
    "id": "term-1764269950037",
    "term": "целеполагание",
    "definition": "A field of study in psychology focused on goal-setting, with several major schools of thought including one from Edwin Locke",
    "aliases": [
      "goal-setting"
    ],
    "firstMentioned": "",
    "frequency": 1,
    "approved": true
  }
]

{
  "name": "transcript-analyzer",
  "version": "0.1.0",
  "private": true,
  "scripts": {
    "dev": "next dev",
    "build": "next build",
    "start": "next start",
    "lint": "eslint",
    "cli": "tsx cli.ts"
  },
  "dependencies": {
    "@anthropic-ai/sdk": "^0.71.0",
    "@radix-ui/react-dialog": "^1.1.15",
    "@radix-ui/react-label": "^2.1.8",
    "@radix-ui/react-progress": "^1.1.8",
    "@radix-ui/react-scroll-area": "^1.2.10",
    "@radix-ui/react-separator": "^1.1.8",
    "@radix-ui/react-slot": "^1.2.4",
    "@radix-ui/react-tabs": "^1.1.13",
    "class-variance-authority": "^0.7.1",
    "clsx": "^2.1.1",
    "lucide-react": "^0.555.0",
    "next": "16.0.5",
    "openai": "^6.9.1",
    "react": "19.2.0",
    "react-dom": "19.2.0",
    "tailwind-merge": "^3.4.0"
  },
  "devDependencies": {
    "@tailwindcss/postcss": "^4",
    "@types/node": "^20",
    "@types/react": "^19",
    "@types/react-dom": "^19",
    "eslint": "^9",
    "eslint-config-next": "16.0.5",
    "tailwindcss": "^4",
    "tsx": "^4.16.5",
    "tw-animate-css": "^1.4.0",
    "typescript": "^5"
  }
}

import OpenAI from 'openai';
import { Extraction, ExtractionType, GlossaryTerm } from '@/types';
import { mockProcessTranscript } from './mockExtractor';
import { areTermsEquivalent } from './term-utils';

const cerebrasApiKey = process.env.CEREBRAS_API_KEY;
const cerebras = cerebrasApiKey ? new OpenAI({
  apiKey: cerebrasApiKey,
  baseURL: 'https://api.cerebras.ai/v1'
}) : null;
const forceMock = process.env.MOCK_TRANSCRIPT_PROCESSOR === 'true';

export interface ProcessTranscriptOptions {
  transcript: string;
  glossary?: GlossaryTerm[];
  chunkSize?: number;
  maxTerms?: number;
}

export async function processTranscript({
  transcript,
  glossary = [],
  chunkSize = 3000,
  maxTerms
}: ProcessTranscriptOptions) {
  if (!transcript) {
    throw new Error('Transcript required');
  }

  if (!cerebras || forceMock) {
    const mockResult = mockProcessTranscript(transcript, glossary, maxTerms);
    return {
      ...mockResult,
      stats: {
        chunks: 1,
        totalExtractions: mockResult.extractions.length,
        byType: countByType(mockResult.extractions),
        newTerms: mockResult.suggestedTerms.length,
        mode: 'mock' as const
      }
    };
  }

  const MODEL = 'llama-3.3-70b';
  const chunks = chunkTranscript(transcript, chunkSize);
  const allExtractions: Extraction[] = [];
  const suggestedTerms: GlossaryTerm[] = [];
  let totalInputTokens = 0;
  let totalOutputTokens = 0;

  console.log(`   Split into ${chunks.length} chunk${chunks.length > 1 ? 's' : ''}`);

  for (let i = 0; i < chunks.length; i++) {
    const chunk = chunks[i];
    console.log(`   Processing chunk ${i + 1}/${chunks.length}...`);

    const extractionResponse = await cerebras.chat.completions.create({
      model: MODEL,
      max_tokens: 4096,
      messages: [{
        role: 'user',
        content: buildExtractionPrompt(glossary, chunk, i + 1, chunks.length)
      }]
    });

    totalInputTokens += extractionResponse.usage?.prompt_tokens || 0;
    totalOutputTokens += extractionResponse.usage?.completion_tokens || 0;

    const extractionText = extractionResponse.choices[0]?.message?.content || '';

    const chunkExtractions = parseExtractions(extractionText, i);
    allExtractions.push(...chunkExtractions);
    console.log(`     → Found ${chunkExtractions.length} extractions`);

    const termResponse = await cerebras.chat.completions.create({
      model: MODEL,
      max_tokens: 2048,
      messages: [{
        role: 'user',
        content: buildTermPrompt(chunk)
      }]
    });

    totalInputTokens += termResponse.usage?.prompt_tokens || 0;
    totalOutputTokens += termResponse.usage?.completion_tokens || 0;

    const termText = termResponse.choices[0]?.message?.content || '';

    const chunkTerms = parseTerms(termText, i);
    suggestedTerms.push(...chunkTerms);
    console.log(`     → Found ${chunkTerms.length} terms`);
  }

  const mergedExtractions = deduplicateExtractions(allExtractions);
  const mergedTerms = limitTerms(deduplicateTerms(suggestedTerms, glossary), maxTerms);

  return {
    extractions: mergedExtractions,
    suggestedTerms: mergedTerms,
    stats: {
      chunks: chunks.length,
      totalExtractions: mergedExtractions.length,
      byType: countByType(mergedExtractions),
      newTerms: mergedTerms.length,
      mode: 'live' as const,
      model: MODEL,
      inputTokens: totalInputTokens,
      outputTokens: totalOutputTokens
    }
  };
}

function buildExtractionPrompt(
  glossary: GlossaryTerm[],
  chunk: string,
  index: number,
  total: number
) {
  return `You are analyzing a meeting transcript. Extract the following with confidence scores (0-100):

1. DECISIONS - explicit agreements or choices made (type: "decision")
2. ACTION ITEMS - tasks assigned to people (type: "action")
3. OPINIONS - viewpoints expressed but not agreed upon (type: "opinion")
4. QUESTIONS - unresolved questions raised (type: "question")
5. TERMS - domain-specific terminology that should be in a glossary (type: "term")

For each extraction, provide:
- type: one of decision/action/opinion/question/term
- content: the extracted content
- confidence: 0-100 score (how certain you are this is correctly classified)
- speaker: who said it (if identifiable)
- sourceSnippet: exact quote from transcript (max 200 chars)
- relatedTerms: any domain terms used

GLOSSARY CONTEXT (use these definitions, suggest new terms not in this list):
${formatGlossary(glossary)}

Return JSON array of extractions. Be conservative - only extract clear items with confidence > 60.

TRANSCRIPT CHUNK ${index}/${total}:
${chunk}`;
}

function buildTermPrompt(chunk: string) {
  return `Analyze this transcript chunk for domain-specific terminology that should be in a glossary.

For each term found:
- term: the term or phrase
- definition: inferred definition from context
- aliases: alternative ways it's referred to

Only include terms that are:
1. Domain-specific (not common words)
2. Used multiple times OR defined in context
3. Would help someone new understand the discussion

Return JSON array of terms.

TRANSCRIPT CHUNK:
${chunk}`;
}

function chunkTranscript(text: string, maxChars: number): string[] {
  const chunks: string[] = [];
  const lines = text.split('\n');
  let currentChunk = '';

  for (const line of lines) {
    if (currentChunk.length + line.length > maxChars && currentChunk.length > 0) {
      chunks.push(currentChunk.trim());
      currentChunk = line;
    } else {
      currentChunk += '\n' + line;
    }
  }

  if (currentChunk.trim()) {
    chunks.push(currentChunk.trim());
  }

  return chunks;
}

function formatGlossary(terms: GlossaryTerm[]): string {
  if (terms.length === 0) return 'No existing glossary terms.';
  return terms.map(t => `- ${t.term}: ${t.definition}`).join('\n');
}

function parseExtractions(text: string, chunkIndex: number): Extraction[] {
  try {
    const jsonMatch = text.match(/\[[\s\S]*\]/);
    if (!jsonMatch) return [];

    const parsed = JSON.parse(jsonMatch[0]);
    return parsed.map((item: Record<string, unknown>, idx: number) => ({
      id: `ext-${chunkIndex}-${idx}`,
      type: (item.type as ExtractionType) || 'opinion',
      content: String(item.content || ''),
      confidence: Number(item.confidence) || 50,
      speaker: item.speaker ? String(item.speaker) : undefined,
      sourceSnippet: String(item.sourceSnippet || item.source_snippet || ''),
      relatedTerms: Array.isArray(item.relatedTerms) ? item.relatedTerms : []
    }));
  } catch {
    return [];
  }
}

function parseTerms(text: string, chunkIndex: number): GlossaryTerm[] {
  try {
    const jsonMatch = text.match(/\[[\s\S]*\]/);
    if (!jsonMatch) return [];

    const parsed = JSON.parse(jsonMatch[0]);
    return parsed.map((item: Record<string, unknown>, idx: number) => ({
      id: `term-${chunkIndex}-${idx}`,
      term: String(item.term || ''),
      definition: String(item.definition || ''),
      aliases: Array.isArray(item.aliases) ? item.aliases : [],
      firstMentioned: '',
      frequency: 1,
      approved: false
    }));
  } catch {
    return [];
  }
}

function deduplicateExtractions(extractions: Extraction[]): Extraction[] {
  const seen = new Map<string, Extraction>();

  for (const ext of extractions) {
    const key = `${ext.type}-${ext.content.toLowerCase().slice(0, 50)}`;
    const existing = seen.get(key);

    if (!existing || ext.confidence > existing.confidence) {
      seen.set(key, ext);
    }
  }

  return Array.from(seen.values()).sort((a, b) => b.confidence - a.confidence);
}

function deduplicateTerms(terms: GlossaryTerm[], existingGlossary: GlossaryTerm[]): GlossaryTerm[] {
  const seenLabels: string[] = [];
  existingGlossary.forEach(term => {
    seenLabels.push(term.term);
    term.aliases?.forEach(alias => seenLabels.push(alias));
  });

  const deduped: GlossaryTerm[] = [];

  for (const term of terms) {
    if (!term.term?.trim()) continue;

    if (seenLabels.some(label => areTermsEquivalent(label, term.term))) {
      continue;
    }

    term.frequency = term.frequency ?? 1;

    const existing = deduped.find(t => areTermsEquivalent(t.term, term.term));
    if (existing) {
      existing.frequency += term.frequency;
      continue;
    }

    deduped.push(term);
    seenLabels.push(term.term);
    term.aliases?.forEach(alias => seenLabels.push(alias));
  }

  return deduped.sort((a, b) => b.frequency - a.frequency);
}

function countByType(extractions: Extraction[]): Record<string, number> {
  const counts: Record<string, number> = {};
  for (const ext of extractions) {
    counts[ext.type] = (counts[ext.type] || 0) + 1;
  }
  return counts;
}

function limitTerms(terms: GlossaryTerm[], maxTerms?: number) {
  if (typeof maxTerms !== 'number' || maxTerms <= 0) {
    return terms;
  }
  return terms.slice(0, maxTerms);
}

import { Extraction, GlossaryTerm, ProcessingStats } from '@/types';

export function buildExtractionsMarkdown(extractions: Extraction[]): string {
  if (extractions.length === 0) {
    return '# Extractions\n\nNo extractions available.';
  }

  const grouped = extractions.reduce<Record<string, Extraction[]>>((acc, extraction) => {
    const key = extraction.type;
    acc[key] = acc[key] || [];
    acc[key].push(extraction);
    return acc;
  }, {});

  const sections = Object.entries(grouped)
    .sort(([a], [b]) => a.localeCompare(b))
    .map(([type, items]) => {
      const header = `## ${capitalize(type)}`;
      const body = items
        .map((item, idx) => {
          const speaker = item.speaker ? `**Speaker:** ${item.speaker}\n` : '';
          const terms = item.relatedTerms && item.relatedTerms.length > 0
            ? `**Related Terms:** ${item.relatedTerms.join(', ')}\n`
            : '';
          const source = item.sourceSnippet ? `> ${item.sourceSnippet}` : '';
          return `### ${idx + 1}. ${item.content}\n${speaker}${terms}**Confidence:** ${item.confidence}%\n${source}`;
        })
        .join('\n\n');
      return `${header}\n\n${body}`;
    });

  return ['# Extractions', ...sections].join('\n\n');
}

export function buildGlossaryMarkdown(
  approved: GlossaryTerm[],
  suggested: GlossaryTerm[] = []
): string {
  const approvedSection = approved.length > 0
    ? approved
        .map((term, idx) => formatTerm(idx + 1, term))
        .join('\n\n')
    : 'No approved terms available.';

  const suggestedSection = suggested.length > 0
    ? suggested
        .map((term, idx) => formatTerm(idx + 1, term))
        .join('\n\n')
    : 'No suggested terms available.';

  return [
    '# Glossary',
    '## Approved Terms',
    approvedSection,
    '## Suggested Terms',
    suggestedSection
  ].join('\n\n');
}

export function buildCombinedMarkdown(
  extractions: Extraction[],
  approved: GlossaryTerm[],
  suggested: GlossaryTerm[] = []
): string {
  const parts = [
    buildExtractionsMarkdown(extractions),
    buildGlossaryMarkdown(approved, suggested)
  ];
  return parts.join('\n\n---\n\n');
}

export interface FullDocumentOptions {
  transcript: string;
  extractions: Extraction[];
  glossaryTerms: GlossaryTerm[];
  suggestedTerms: GlossaryTerm[];
  stats: ProcessingStats;
  includeTranscript?: boolean;
  includeExtractions?: boolean;
  includeGlossary?: boolean;
}

export function buildFullDocument(options: FullDocumentOptions): string {
  const {
    transcript,
    extractions,
    glossaryTerms,
    suggestedTerms,
    stats,
    includeTranscript = false,
    includeExtractions = true,
    includeGlossary = true
  } = options;

  const parts: string[] = [];

  // Metadata section
  parts.push(buildMetadataSection(stats));

  // Optional transcript section
  if (includeTranscript) {
    parts.push('# Transcript\n\n' + transcript);
  }

  // Optional extractions section
  if (includeExtractions && extractions.length > 0) {
    parts.push(buildExtractionsMarkdown(extractions));
  }

  // Optional glossary section
  if (includeGlossary && (glossaryTerms.length > 0 || suggestedTerms.length > 0)) {
    parts.push(buildGlossaryMarkdown(glossaryTerms, suggestedTerms));
  }

  return parts.join('\n\n---\n\n');
}

function buildMetadataSection(stats: ProcessingStats): string {
  const lines = [
    '---',
    `chunks: ${stats.chunks}`,
    `extractions: ${stats.totalExtractions}`,
    `new_terms: ${stats.newTerms}`,
    `mode: ${stats.mode}`
  ];

  if (stats.model) {
    lines.push(`model: ${stats.model}`);
  }

  if (stats.inputTokens !== undefined) {
    lines.push(`input_tokens: ${stats.inputTokens}`);
  }

  if (stats.outputTokens !== undefined) {
    lines.push(`output_tokens: ${stats.outputTokens}`);
  }

  if (stats.inputTokens && stats.outputTokens) {
    lines.push(`total_tokens: ${stats.inputTokens + stats.outputTokens}`);
  }

  const typeBreakdown = Object.entries(stats.byType)
    .map(([type, count]) => `  ${type}: ${count}`)
    .join('\n');

  if (typeBreakdown) {
    lines.push(`extractions_by_type:\n${typeBreakdown}`);
  }

  lines.push('---');

  return lines.join('\n');
}

function formatTerm(index: number, term: GlossaryTerm) {
  const aliases = term.aliases && term.aliases.length > 0
    ? `**Aliases:** ${term.aliases.join(', ')}\n`
    : '';
  return `### ${index}. ${term.term}\n${term.definition}\n\n${aliases}**First Mentioned:** ${term.firstMentioned || 'n/a'}\n**Frequency:** ${term.frequency}`;
}

function capitalize(value: string) {
  return value.charAt(0).toUpperCase() + value.slice(1);
}

import { Extraction, ExtractionType, GlossaryTerm } from '@/types';
import { areTermsEquivalent, normalizeTermKey } from './term-utils';

export function mockProcessTranscript(
  transcript: string,
  glossary: GlossaryTerm[],
  maxTerms?: number
) {
  const sentences = splitTranscript(transcript);
  const extracted: Extraction[] = [];

  sentences.forEach((sentence, idx) => {
    const parsed = classifySentence(sentence, glossary, idx);
    if (parsed) {
      extracted.push(parsed);
    }
  });

  const suggestedTerms = buildGlossarySuggestions(transcript, glossary, maxTerms);

  return {
    extractions: extracted,
    suggestedTerms
  };
}

function splitTranscript(transcript: string): string[] {
  return transcript
    .split(/\n+|(?<=[.!?])\s+/)
    .map(part => part.trim())
    .filter(Boolean);
}

function classifySentence(
  sentence: string,
  glossary: GlossaryTerm[],
  index: number
): Extraction | null {
  let working = sentence;
  let speaker: string | undefined;

  const speakerMatch = working.match(/^([A-Za-z][A-Za-z\s]{1,30}):\s*(.+)$/);
  if (speakerMatch) {
    speaker = speakerMatch[1].trim();
    working = speakerMatch[2].trim();
  }

  const lowered = working.toLowerCase();

  let type: ExtractionType | null = null;
  let confidence = 65;

  if (/[?？]$/.test(working)) {
    type = 'question';
    confidence = 90;
  } else if (/\b(decision|decide|agreed|agreement|approve)\b/.test(lowered)) {
    type = 'decision';
    confidence = 80;
  } else if (/\b(action item|follow up|will|need to|let's|assign|task)\b/.test(lowered)) {
    type = 'action';
    confidence = 78;
  } else if (/\b(i think|i believe|feels like|i guess|should)\b/.test(lowered)) {
    type = 'opinion';
    confidence = 70;
  }

  if (!type) {
    return null;
  }

  const related = glossary
    .filter(term => containsTerm(lowered, term.term))
    .map(term => term.term);

  return {
    id: `mock-ext-${index}`,
    type,
    content: working,
    confidence,
    speaker,
    sourceSnippet: working.slice(0, 200),
    relatedTerms: related
  };
}

function containsTerm(content: string, term: string) {
  return content.includes(term.toLowerCase());
}

export function buildGlossarySuggestions(
  transcript: string,
  glossary: GlossaryTerm[],
  maxTerms?: number
): GlossaryTerm[] {
  const existingLabels: string[] = [];
  glossary.forEach(term => {
    existingLabels.push(term.term);
    term.aliases?.forEach(alias => existingLabels.push(alias));
  });

  const tokens = transcript.match(/\b[A-Z][A-Za-z0-9]+(?:\s+[A-Z][A-Za-z0-9]+)?\b/g) || [];
  const aggregates: { label: string; count: number }[] = [];

  tokens.forEach(token => {
    const normalized = token.trim();
    if (normalized.length < 3) return;
    if (!normalizeTermKey(normalized)) return;

    if (existingLabels.some(label => areTermsEquivalent(label, normalized))) {
      return;
    }

    const entry = aggregates.find(item => areTermsEquivalent(item.label, normalized));
    if (entry) {
      entry.count += 1;
    } else {
      aggregates.push({ label: normalized, count: 1 });
    }
  });

  const sorted = aggregates.sort((a, b) => b.count - a.count);
  const limit = typeof maxTerms === 'number' && maxTerms > 0 ? maxTerms : 5;

  return sorted.slice(0, limit).map((info, idx) => ({
    id: `mock-term-${idx}`,
    term: info.label,
    definition: deriveDefinition(info.label, transcript),
    aliases: [],
    firstMentioned: 'mock',
    frequency: info.count,
    approved: false
  }));
}

function deriveDefinition(term: string, transcript: string): string {
  const pattern = new RegExp(`[^.!?]*${escapeRegExp(term)}[^.!?]*[.!?]`, 'i');
  const match = transcript.match(pattern);
  if (match) {
    return match[0].trim();
  }
  return `Term mentioned in transcript context as "${term}".`;
}

function escapeRegExp(value: string) {
  return value.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
}

export function normalizeTermKey(value: string): string {
  return value
    .toLowerCase()
    .replace(/[^a-z0-9]/g, '')
    .trim();
}

export function areTermsEquivalent(a: string, b: string): boolean {
  const normA = normalizeTermKey(a);
  const normB = normalizeTermKey(b);

  if (!normA || !normB) return false;
  if (normA === normB) return true;

  return levenshtein(normA, normB) <= 2;
}

function levenshtein(a: string, b: string): number {
  const aLen = a.length;
  const bLen = b.length;

  if (aLen === 0) return bLen;
  if (bLen === 0) return aLen;

  const prev = new Array(bLen + 1).fill(0);
  const curr = new Array(bLen + 1).fill(0);

  for (let j = 0; j <= bLen; j++) {
    prev[j] = j;
  }

  for (let i = 1; i <= aLen; i++) {
    curr[0] = i;

    for (let j = 1; j <= bLen; j++) {
      const cost = a[i - 1] === b[j - 1] ? 0 : 1;
      curr[j] = Math.min(
        prev[j] + 1,
        curr[j - 1] + 1,
        prev[j - 1] + cost
      );
    }

    for (let j = 0; j <= bLen; j++) {
      prev[j] = curr[j];
    }
  }

  return curr[bLen];
}

import { clsx, type ClassValue } from "clsx"
import { twMerge } from "tailwind-merge"

export function cn(...inputs: ClassValue[]) {
  return twMerge(clsx(inputs))
}

export type ExtractionType = 'decision' | 'action' | 'opinion' | 'question' | 'term';

export interface Extraction {
  id: string;
  type: ExtractionType;
  content: string;
  confidence: number; // 0-100
  speaker?: string;
  sourceSnippet: string;
  sourceTimestamp?: string;
  relatedTerms?: string[];
}

export interface GlossaryTerm {
  id: string;
  term: string;
  definition: string;
  aliases: string[];
  firstMentioned: string; // meeting ID
  frequency: number;
  approved: boolean;
}

export interface TranscriptMeeting {
  id: string;
  title: string;
  date: string;
  filePath: string;
  extractions: Extraction[];
  suggestedTerms: GlossaryTerm[];
}

export interface ProcessingResult {
  meeting: TranscriptMeeting;
  glossaryUpdates: GlossaryTerm[];
}

export interface ProcessingStats {
  chunks: number;
  totalExtractions: number;
  byType: Record<string, number>;
  newTerms: number;
  mode: 'live' | 'mock';
  inputTokens?: number;
  outputTokens?: number;
  model?: string;
}

{
  "compilerOptions": {
    "target": "ES2017",
    "lib": ["dom", "dom.iterable", "esnext"],
    "allowJs": true,
    "skipLibCheck": true,
    "strict": true,
    "noEmit": true,
    "esModuleInterop": true,
    "module": "esnext",
    "moduleResolution": "bundler",
    "resolveJsonModule": true,
    "isolatedModules": true,
    "jsx": "react-jsx",
    "incremental": true,
    "plugins": [
      {
        "name": "next"
      }
    ],
    "paths": {
      "@/*": ["./src/*"]
    }
  },
  "include": [
    "next-env.d.ts",
    "**/*.ts",
    "**/*.tsx",
    ".next/types/**/*.ts",
    ".next/dev/types/**/*.ts",
    "**/*.mts"
  ],
  "exclude": ["node_modules"]
}