Follow-up query rewrite

Source: platform/apps/app/src/lib/rag/query-rewrite.ts · rendered from main on every deploy — edit in the repo, not here

platform/apps/app/src/lib/rag/query-rewrite.ts

/**
 * Query rewriting for multi-turn chat (ADR-0017 §2).
 *
 * Takes conversation history + current query and rewrites the query
 * as a standalone question that captures all necessary context.
 * This enables the existing RAG pipeline to work without modification.
 */

import { generateText } from 'ai';

interface HistoryMessage {
  role: 'user' | 'assistant';
  content: string;
}

interface RewriteResult {
  rewritten: string;
  wasRewritten: boolean;
}

/**
 * Rewrite a follow-up query into a standalone question using conversation history.
 * Returns the original query unchanged if:
 * - No history exists (first turn)
 * - The query is already self-contained (heuristic)
 * - LLM call fails (graceful fallback)
 */
export async function rewriteQuery(
  history: HistoryMessage[],
  currentQuery: string,
  apiKey: string,
  locale: string,
  gw?: { AI_GATEWAY_URL?: string; AI_GATEWAY_TOKEN?: string; ANTHROPIC_API_KEY?: string },
): Promise<RewriteResult> {
  // First turn or very short history — no rewrite needed
  if (history.length === 0) {
    return { rewritten: currentQuery, wasRewritten: false };
  }

  // Heuristic: skip rewrite for trivial follow-ups that don't reference context
  const referencePatterns =
    locale === 'th'
      ? /(?:นั้น|นี้|มัน|เรื่องนี้|ข้อนั้น|ข้างต้น|ดังกล่าว|อันไหน|ตรงนั้น|แล้ว.*ล่ะ|เพิ่มเติม)/
      : /(?:\b(?:it|this|that|these|those|they|them|its|their|above|previous|mentioned|said|the same|more about|what about|how about)\b)/i;

  // Short queries without references are standalone (e.g., "ค่าธรรมเนียม", "ISO 9001")
  if (!referencePatterns.test(currentQuery) && currentQuery.length > 15) {
    return { rewritten: currentQuery, wasRewritten: false };
  }

  // Very short acknowledgements — no rewrite value
  const trivialPatterns = /^(ใช่|ไม่|ครับ|ค่ะ|ok|yes|no|thanks|ขอบคุณ|ดี|โอเค)\s*[.!?]*$/i;
  if (trivialPatterns.test(currentQuery.trim())) {
    return { rewritten: currentQuery, wasRewritten: false };
  }

  // Build compact history (last 6 messages, truncated)
  const recentHistory = history.slice(-6).map((m) => ({
    role: m.role,
    content: m.content.slice(0, 300),
  }));

  const historyText = recentHistory
    .map((m) => `${m.role === 'user' ? 'User' : 'Assistant'}: ${m.content}`)
    .join('\n');

  const systemPrompt =
    locale === 'th'
      ? `เขียนคำถามของผู้ใช้ใหม่ให้เป็นคำถามที่สมบูรณ์ในตัวเอง โดยรวมบริบทจากประวัติสนทนา ตอบเฉพาะคำถามที่เขียนใหม่เท่านั้น ไม่ต้องอธิบายเพิ่ม`
      : `Rewrite the user's latest question as a standalone query that captures all necessary context from the conversation history. Output only the rewritten query, nothing else.`;

  try {
    const { createLlmProvider } = await import('../chat/llm-provider.js');
    const { chatModel } = createLlmProvider(apiKey, gw);

    const { text } = await generateText({
      model: chatModel,
      system: systemPrompt,
      prompt: `Conversation:\n${historyText}\n\nLatest question: ${currentQuery}`,
      maxOutputTokens: 150,
      temperature: 0,
    });

    const rewritten = text.trim();
    if (!rewritten || rewritten.length < 3) {
      return { rewritten: currentQuery, wasRewritten: false };
    }

    return { rewritten, wasRewritten: true };
  } catch {
    // LLM failed — use original query (graceful degradation)
    return { rewritten: currentQuery, wasRewritten: false };
  }
}