receipt-parser/src/index.mjs

#!/usr/bin/env node
"use strict";

import path from 'path';
import fs from 'fs';

import tqdm from 'tqdm';
import pThrottle from 'p-throttle';

import ai_extract from './lib/ai-extract.mjs';
import make_html from './lib/make-html.mjs';

// HACK: Make sure __dirname is defined when using es6 modules. I forget where I found this - a PR with a source URL would be great!
const __dirname = import.meta.url.slice(7, import.meta.url.lastIndexOf("/"));

// const filepaths = process.argv.slice(2).map(f => path.resolve(process.cwd(), f));
const filepaths = fs.readFileSync(process.argv.slice(2)[0], "utf-8")
	.split(`\n`)
	.map(el => el.trim())
	.filter(el => el.length > 0);

console.log(`Hello from Node.js`);

console.log(`>>> FILEPATHS`);
console.log(filepaths.join(`\n`));

const txts = await Promise.all(filepaths.map(filepath => fs.promises.readFile(filepath, "utf-8")));


const throttler = pThrottle({
	limit: 40,
	interval: 60 * 1000 // every minute
});
const throttled = throttler(ai_extract);

const objs = [];
let i = 0;
for(let txt of tqdm(txts, { total: txts.length })) {
	console.log(`>>> AI > ${i}`);
	
	txt = txt.trim();
	if(txt.length === 0) {
		objs.push({
			date: "unknown",
			item_name: "unknown",
			paid: null,
			vat: null,
			category: "Unknown"
		});
		continue;
	}
	
	const result = await throttled(txt, false); // bool is whether we're pretend or not - i.e. not making anthropic/claude api calls
	result.i = i;
	
	console.log(result);
	
	objs.push(result);
	
	i++;
}

const html = make_html(objs);
const filepath_out = path.join(process.cwd(), `report.html`);
fs.writeFileSync(filepath_out, html);

console.log(`Written output to ${filepath_out}`);