import { Pinecone } from "@pinecone-database/pinecone";
import { getEnvOrThrow } from "@/lib/utils";
import { embed } from "ai";
import { createOpenAI } from "@ai-sdk/openai";

const PINECONE_THRESHOLD_VECTORS = 20

const openai = createOpenAI({ apiKey: process.env.OPENAI_API_KEY ?? "" });

export async function retrieveRelevantSnippets(
  query: string,
  topK: number
): Promise<string[]> {
  const pc = new Pinecone({ apiKey: getEnvOrThrow("PINECONE_API_KEY") });
  const index = pc.Index(getEnvOrThrow("PINECONE_INDEX"));

  const { embedding } = await embed({
    model: openai.embedding("text-embedding-ada-002"),
    value: "All of the products mentioned in this description:\n" + query,
  });

  const result = await index.query({
    vector: embedding,
    topK: topK + PINECONE_THRESHOLD_VECTORS,
    includeMetadata: true,
  });

  type Match = {
    metadata?: {
      text?: string;
      code?: string | number;
      desc?: string;
      gdesc?: string;
      category?: string;
      [key: string]: unknown;
    };
  };
  const matches = (result.matches ?? []) as unknown as Match[];
  const snippets = matches
    .map((m) => {
      if (!m.metadata) return "";
      const lines: string[] = [];
      if (m.metadata.desc) lines.push(`The Item: ${m.metadata.desc}`);
      if (m.metadata.gdesc)
        lines.push(`has Goederen Omschrijving: ${m.metadata.gdesc}`);
      if (m.metadata.code !== undefined)
        lines.push(`and GOEDEREN CODE: ${String(m.metadata.code)}`);
      // if (m.metadata.category) lines.push(`Category: ${m.metadata.category}`)
      // Include any generic text field last if present
      // if (m.metadata.text) lines.push(`Text: ${m.metadata.text}`)
      return lines.join("\n");
    })
    .filter(Boolean);
  return snippets;
}