# frozen_string_literal: true

# Embedder — turns text into vectors via OpenRouter's /embeddings endpoint (OpenAI-compatible),
# reusing the same key + HTTP/auth pattern as the summariser. Standard library only. The HTTP transport is
# injectable so the RAG index/retrieval is unit-testable offline (no network/key).

require "json"
require "net/http"
require "uri"

module Helpdesk
  class Embedder
    MODEL = "openai/text-embedding-3-small" # 1536-dim, ~$0.02/M tokens, handles Czech
    URL = "https://openrouter.ai/api/v1/embeddings"

    def initialize(api_key: ENV["OPENROUTER_API_KEY"], model: MODEL, transport: nil)
      @api_key = api_key
      @model = model
      @transport = transport # ->(request_body_json) { response_body_json } — for tests
    end

    # texts: String | Array<String>  →  Array<Array<Float>> (order preserved). Raises on API error.
    def embed(texts)
      arr = texts.is_a?(Array) ? texts : [texts]
      return [] if arr.empty?
      body = JSON.generate(model: @model, input: arr)
      raw = @transport ? @transport.call(body) : post(body)
      data = JSON.parse(raw)
      raise "embedding error: #{data["error"]}" if data["error"]
      data.fetch("data").sort_by { |d| d["index"] }.map { |d| d.fetch("embedding") }
    end

    def embed_one(text) = embed(text).first

    private

    def post(body)
      uri = URI(URL)
      http = Net::HTTP.new(uri.host, uri.port)
      http.use_ssl = true
      http.open_timeout = 15
      http.read_timeout = 60
      req = Net::HTTP::Post.new(uri)
      req["Authorization"] = "Bearer #{@api_key}"
      req["Content-Type"] = "application/json"
      req.body = body
      http.request(req).body
    end
  end
end