Add R personal shopper using OpenAI/Ollama + SerpAPI

interests_sample.txt

@@ -0,0 +1,4 @@
+# One item per line
+wireless noise cancelling headphones
+ergonomic office chair
+espresso machine

personal_shopper.R

@@ -0,0 +1,295 @@
+#!/usr/bin/env Rscript
+
+suppressPackageStartupMessages({
+  library(httr2)
+  library(jsonlite)
+  library(dplyr)
+  library(purrr)
+  library(readr)
+  library(stringr)
+  library(tibble)
+})
+
+# -------------------------
+# Configuration
+# -------------------------
+DEFAULT_ENGINES <- c("google_shopping", "bing_shopping", "ebay")
+
+read_interest_items <- function(path) {
+  if (!file.exists(path)) {
+    stop(sprintf("Input file not found: %s", path))
+  }
+
+  lines <- read_lines(path, lazy = FALSE, progress = FALSE)
+  items <- lines |>
+    str_trim() |>
+    discard(~ .x == "" || str_starts(.x, "#"))
+
+  if (length(items) == 0) {
+    stop("No shopping items found in input file.")
+  }
+
+  items
+}
+
+safe_num <- function(x) {
+  if (is.null(x) || length(x) == 0 || is.na(x)) return(NA_real_)
+  out <- str_extract(as.character(x), "[0-9]+(?:\\.[0-9]+)?")
+  as.numeric(out)
+}
+
+openai_chat_json <- function(messages, model = "gpt-4.1-mini") {
+  api_key <- Sys.getenv("OPENAI_API_KEY")
+  if (api_key == "") {
+    stop("OPENAI_API_KEY is required for provider=openai")
+  }
+
+  body <- list(
+    model = model,
+    messages = messages,
+    response_format = list(type = "json_object"),
+    temperature = 0.2
+  )
+
+  resp <- request("https://api.openai.com/v1/chat/completions") |>
+    req_headers(Authorization = paste("Bearer", api_key)) |>
+    req_body_json(body, auto_unbox = TRUE) |>
+    req_perform()
+
+  parsed <- resp_body_json(resp, simplifyVector = TRUE)
+  parsed$choices[[1]]$message$content
+}
+
+ollama_chat_json <- function(messages, model = "llama3.1") {
+  host <- Sys.getenv("OLLAMA_HOST", "http://localhost:11434")
+  body <- list(
+    model = model,
+    messages = messages,
+    format = "json",
+    stream = FALSE,
+    options = list(temperature = 0.2)
+  )
+
+  resp <- request(paste0(host, "/api/chat")) |>
+    req_body_json(body, auto_unbox = TRUE) |>
+    req_perform()
+
+  parsed <- resp_body_json(resp, simplifyVector = TRUE)
+  parsed$message$content
+}
+
+llm_chat_json <- function(provider, model, messages) {
+  if (provider == "openai") {
+    openai_chat_json(messages, model)
+  } else if (provider == "ollama") {
+    ollama_chat_json(messages, model)
+  } else {
+    stop("provider must be either 'openai' or 'ollama'")
+  }
+}
+
+generate_queries <- function(item, provider, model) {
+  prompt <- paste0(
+    "Create exactly 3 short shopping search queries for this item: ", item,
+    ". Return strict JSON: {\"queries\":[\"...\",\"...\",\"...\"]}."
+  )
+
+  content <- llm_chat_json(
+    provider = provider,
+    model = model,
+    messages = list(
+      list(role = "system", content = "You are a shopping query optimizer."),
+      list(role = "user", content = prompt)
+    )
+  )
+
+  parsed <- fromJSON(content)
+  queries <- unique(unlist(parsed$queries))
+  queries <- queries[!is.na(queries) & queries != ""]
+
+  if (length(queries) == 0) {
+    c(item, paste(item, "best deal"), paste(item, "buy online"))
+  } else {
+    queries
+  }
+}
+
+search_serpapi <- function(query, engine, serpapi_key, max_results = 10) {
+  resp <- request("https://serpapi.com/search.json") |>
+    req_url_query(
+      q = query,
+      engine = engine,
+      api_key = serpapi_key,
+      num = max_results
+    ) |>
+    req_perform()
+
+  data <- resp_body_json(resp, simplifyVector = TRUE)
+
+  if (!is.null(data$shopping_results)) {
+    results <- data$shopping_results
+    return(tibble(
+      title = results$title %||% NA_character_,
+      price_text = results$price %||% NA_character_,
+      price = map_dbl(results$price, safe_num),
+      store = results$source %||% NA_character_,
+      link = results$link %||% NA_character_,
+      engine = engine,
+      query = query
+    ))
+  }
+
+  if (!is.null(data$organic_results)) {
+    results <- data$organic_results
+    return(tibble(
+      title = results$title %||% NA_character_,
+      price_text = NA_character_,
+      price = NA_real_,
+      store = results$source %||% engine,
+      link = results$link %||% NA_character_,
+      engine = engine,
+      query = query
+    ))
+  }
+
+  tibble(
+    title = character(),
+    price_text = character(),
+    price = numeric(),
+    store = character(),
+    link = character(),
+    engine = character(),
+    query = character()
+  )
+}
+
+`%||%` <- function(x, y) {
+  if (is.null(x)) y else x
+}
+
+rank_suggestions <- function(item, suggestions, provider, model) {
+  shortlist <- suggestions |>
+    arrange(is.na(price), price) |>
+    slice_head(n = min(8, n())) |>
+    mutate(row_id = row_number())
+
+  if (nrow(shortlist) == 0) {
+    return(suggestions |> mutate(ai_pick = FALSE, ai_note = NA_character_))
+  }
+
+  choices_json <- toJSON(shortlist |> select(row_id, title, price_text, store, link), auto_unbox = TRUE)
+
+  prompt <- paste0(
+    "Choose up to 3 best shopping options for '", item, "' balancing low price and reputable store. ",
+    "Return JSON: {\"picked_row_ids\":[1,2],\"note\":\"...\"}. Options: ", choices_json
+  )
+
+  content <- llm_chat_json(
+    provider = provider,
+    model = model,
+    messages = list(
+      list(role = "system", content = "You are a careful shopping assistant."),
+      list(role = "user", content = prompt)
+    )
+  )
+
+  parsed <- fromJSON(content)
+  picked <- as.integer(unlist(parsed$picked_row_ids))
+  note <- as.character(parsed$note %||% "")
+
+  shortlist <- shortlist |>
+    mutate(ai_pick = row_id %in% picked, ai_note = if_else(ai_pick, note, NA_character_)) |>
+    select(-row_id)
+
+  leftovers <- anti_join(suggestions, shortlist, by = c("title", "price_text", "price", "store", "link", "engine", "query")) |>
+    mutate(ai_pick = FALSE, ai_note = NA_character_)
+
+  bind_rows(shortlist, leftovers)
+}
+
+shop_item <- function(item, provider, model, engines, serpapi_key) {
+  queries <- tryCatch(
+    generate_queries(item, provider, model),
+    error = function(e) {
+      message(sprintf("LLM query generation failed for '%s': %s", item, e$message))
+      c(item, paste(item, "best deal"), paste(item, "buy online"))
+    }
+  )
+
+  results <- map_dfr(engines, function(engine) {
+    map_dfr(queries, function(query) {
+      tryCatch(
+        search_serpapi(query, engine, serpapi_key),
+        error = function(e) {
+          message(sprintf("Search failed for engine=%s query='%s': %s", engine, query, e$message))
+          tibble(
+            title = character(), price_text = character(), price = numeric(),
+            store = character(), link = character(), engine = character(), query = character()
+          )
+        }
+      )
+    })
+  })
+
+  cleaned <- results |>
+    filter(!is.na(title), title != "") |>
+    mutate(item = item) |>
+    distinct(item, title, store, price_text, link, .keep_all = TRUE)
+
+  rank_suggestions(item, cleaned, provider, model)
+}
+
+run_personal_shopper <- function(input_file,
+                                 output_csv = "shopping_suggestions.csv",
+                                 provider = c("openai", "ollama"),
+                                 model = NULL,
+                                 engines = DEFAULT_ENGINES) {
+  provider <- match.arg(provider)
+
+  if (is.null(model)) {
+    model <- if (provider == "openai") "gpt-4.1-mini" else "llama3.1"
+  }
+
+  serpapi_key <- Sys.getenv("SERPAPI_KEY")
+  if (serpapi_key == "") {
+    stop("SERPAPI_KEY is required for search engine shopping lookups.")
+  }
+
+  items <- read_interest_items(input_file)
+
+  all_results <- map_dfr(items, function(item) {
+    message(sprintf("Scouting: %s", item))
+    shop_item(item, provider, model, engines, serpapi_key)
+  }) |>
+    arrange(item, is.na(price), price)
+
+  write_csv(all_results, output_csv)
+
+  cat("\n=== Shopping Suggestions ===\n")
+  print(all_results |>
+          select(item, title, price_text, price, store, engine, ai_pick, ai_note, link), n = 100)
+
+  invisible(all_results)
+}
+
+# CLI usage:
+# Rscript personal_shopper.R interests.txt output.csv openai gpt-4.1-mini
+args <- commandArgs(trailingOnly = TRUE)
+if (sys.nframe() == 0) {
+  if (length(args) < 1) {
+    stop("Usage: Rscript personal_shopper.R <input_txt> [output_csv] [provider=openai|ollama] [model]")
+  }
+
+  input_file <- args[[1]]
+  output_csv <- ifelse(length(args) >= 2, args[[2]], "shopping_suggestions.csv")
+  provider <- ifelse(length(args) >= 3, args[[3]], "openai")
+  model <- ifelse(length(args) >= 4, args[[4]], NA_character_)
+  if (is.na(model)) model <- NULL
+
+  run_personal_shopper(
+    input_file = input_file,
+    output_csv = output_csv,
+    provider = provider,
+    model = model
+  )
+}