workflow on S Anand

Organizing PDF receipts

Thu, 12 Feb 2026 17:21:14 +0800

One of my goals this year is to “Automate finance + tax”. Today, I took a baby step by organizing my expenses.

This is my current process:

STEP 1: Download PDF receipts (from OpenAI, Anthropic, Google, and other cloud/AI services)
STEP 2: Organize them, so I know which receipt to upload against which expense
STEP 3: Submit on SAP Concur.

All steps are manual as of now. I automated STEP 2: Organize them.

After downloading my PDF receipts, I used this set of prompts on Codex - GPT 5.3 Codex (High)

This folder contains a set of PDF receipts.

Write a well-documented Python script `rename_receipts.py` that renames each .pdf into this format: `2026-02-11 Github $20.00 Card-1004.pdf`. Specifically, `YYYY-MM-DD [Service] [Amount] Card-[Last 4 Digits].pdf`.

- The date should be the PAID date, not the invoice date.
- The service should be a single word vendor name, i.e. Github, OpenAI, Anthropic, Weaviate, Cloudflare, Hetzner, nothing else. You only need to process these formats for now.
- The amount should be the total charged, two decimals, with the currency symbol (e.g. $20.00)
- Include the last 4 digits of the card used for payment.

If the target file exists

- If it is identical to the source file, delete the source file.
- If it is different, print a warning and skip renaming.

Run and test.

---

Fall back to the invoice date only if the date of payment is not available. Skip Card-xxxx if unavailable.

---

How is '2025-12-02 Weaviate $25.00 Card-1004.pdf' different from 'Receipt-2796-1548.pdf'?

---

Let's just check the file sizes, then.

---

Simplify the code further where there are clear opportunities. Retain the documentation and improve readability.

---

Skip processing files that already begin with YYYY-MM-DD.

---

I've added three Google receipts. Update the code to process them the same way and run it. The service name to use is "Google".

---

I've added a new Google PDF. Include that as well, with the service name as Google.

---

I added 2 Hostgator PDFs - h1.pdf and h2.pdf. Incorporate these as well.

---

Any clean-up to the script that might help?

---

Yes, apply these. Also, '2026-01-01 Google $0.80 Card-xxxx.pdf' should be '2026-01-31 Google $0.80 Card-xxxx.pdf' -- we want to go with the statement date.

The result is rename_receipts.py and it works well enough for my receipts over the last 3-4 months.

I’ve partly automated STEP 1 as well. The trickiest bit was converting Google Pay Activity into PDF receipts. This is what it looks like>

To convert the right side into PDF receipts, I asked Claude which gave me this script:

(async function() {
  const element = $0.closest(".b3-section-outer-content");
  const doc = element.ownerDocument;
  const win = doc.defaultView || doc.parentWindow;

  // Collect all CSS
  let css = "";
  for (const sheet of doc.styleSheets) {
    try {
      css += Array.from(sheet.cssRules).map(r => r.cssText).join("\n");
    } catch (e) {
      if (sheet.href) css += `@import url('${sheet.href}');\n`;
    }
  }

  // If in iframe, try to get parent styles too
  if (win !== window) {
    try {
      for (const sheet of window.document.styleSheets) {
        try {
          css += Array.from(sheet.cssRules).map(r => r.cssText).join("\n");
        } catch (e) {
          if (sheet.href) css += `@import url('${sheet.href}');\n`;
        }
      }
    } catch (e) {}
  }

  const html = `<!DOCTYPE html>
<html>
<head>
  <meta charset="UTF-8">
  <style>${css}</style>
  <style>
    body { margin: 0; padding: 20px; }
  </style>
</head>
<body>${element.outerHTML}</body>
</html>`;

  const blob = new Blob([html], { type: "text/html;charset=utf-8" });
  const url = URL.createObjectURL(blob);
  const newWin = window.open(url, "_blank");

  if (newWin) {
    newWin.addEventListener("load", () => {
      setTimeout(() => {
        newWin.print();
      }, 500);
    });
  } else {
    console.error("❌ Popup blocked. Please allow popups for this site.");
  }
})();

Now, I can right-click any receipt in Google Pay Activity, select “Inspect”, then run this script in the console. It opens a new tab with just that receipt, which I print to PDF.

The next step is to automate STEP 3: the Concur submission itself. But that’s for next quarter!

From Calvin & Hobbes to Photo Tagging: Excel's Unexpected Image Capability

Tue, 26 Mar 2024 03:25:30 +0000

In Excel, using Visual Basic, you can change an image as you scroll. This makes it easy to look at each image and annotate it.

This is how I transcribed every Calvin & Hobbes.

I used this technique first when typing out the strips during my train rides from Bandra to Churchgate. I had an opportunity to re-apply it recently when we needed to tag hundreds of photographs based on a set of criteria.

Here’s how you can do this. Note: This works only on Windows.

STEP 1: Create a new Excel workbook and save it as an Excel macro-enabled workbook. (Note: When opening it again, you need to enable macros)

STEP 2: Open File > Options (Alt-F-T), go to Customize Ribbon. Under “Customize the Ribbon”, enable the “Developer” menu.

STEP 3: In Developer > Insert > ActiveX Controls, select Image and draw a rectangle from A1 to J10. (Resize it later.)

STEP 4: By default, this will be called Image1. In any case, note down the name from the Name box on the top left.

STEP 5: In cells A11 onwards, add paths to file names.

STEP 6: Click Developer > Visual Basic (Alt-F11), go to ThisWorkbook, and paste this code:

Option Explicit

Private Sub Workbook_SheetSelectionChange(ByVal Sh As Object, ByVal Target As Excel.Range)
    Dim img As String
    img = Sh.Cells(Target.Row, 1).Value
    If (img <> "" And img <> "file") Then ActiveSheet.Image1.Picture = LoadPicture(img)
End Sub

Replace ActiveSheet.Image1 with ActiveSheet.(whatever) based on your image name in Step 4.

STEP 7: Select Developer > Design Mode. Click on Image1. Then select Developer > Properties. In this panel, under PictureSizeMode, choose 3 - fmPictureSizeModeZoom to fit the picture.

Now scroll through the rows. The images will change.

Comments

Karthik 26 Mar 2024 9:54 am: Everything okay, but you were able to SIT and work on a laptop on train rides from Bandra to Churchgate??
S Anand 26 Mar 2024 12:16 pm: Ah, I took the fast trains that started and ended at Bandra and Churchgate 🙂

Cyborg scraping

Fri, 25 Jun 2021 04:03:23 +0000

LinkedIn has a page that shows the people who most recently followed you.

At first, it shows just 20 people. But as you scroll, it keeps fetching the rest. I’d love to get the full list on a spreadsheet. I’m curious about:

What kind of people follow me?
Which of them has the most followers?
Who are my earliest followers?

But first, I need to scrape this list. Normally, I’d spend a day writing a program. But I tried a different approach yesterday.

Aside: it’s easy to get bored in online meetings. I have a surplus of partially distracted time. So rather than writing code to save me time, I’d rather create simple tasks to keep me occupied. Like scrolling.

So here’s my workflow to scrape the list of followers.

Step 1: Keep scrolling all the way to the bottom until you get all followers.

Step 2: Press F12, open the Developer Tools - Console, and paste this code.

copy(
  $$(".follows-recommendation-card").map(v => {
    let name = v.querySelector(".follows-recommendation-card__name");
    let headline = v.querySelector(".follows-recommendation-card__headline");
    let subtext = v.querySelector(".follows-recommendation-card__subtext");
    let link = v.querySelector(".follows-recommendation-card__avatar-link");
    let followers = "", match;
    if (subtext) {
      if (match = subtext.innerText.match(/([\d\.K]+) follower/)) {
        followers = match[1];
      } else if (match = subtext.innerText.match(/([\d\.K]+) other/)) {
        followers = match[1];
      }
    }
    followers = followers.match(/K$/)
      ? parseFloat(followers) * 1000
      : parseFloat(followers);
    return {
      name: name ? name.innerText : "",
      headline: headline ? headline.innerText : "",
      followers: followers,
      link: link ? link.href : "",
    };
  }),
);

Step 3: The name, headline, followers and link are now in the clipboard as JSON. Visit https://www.convertcsv.com/json-to-csv.htm and paste it in “Select your input” under “Enter Data”.

Step 4: Click on the “Download Result” button. The JSON is converted into a CSV you can load into a spreadsheet.

I call this “Cyborg scraping”. I do half the work (scrolling, copy-pasting, etc.) The code does half the work. It’s manual. It’s a bit slow. But it gets the job done quick and dirty.

I’ll share later what I learned about my followers. For now, I’m looking forward to meetings 😉

PS: A similar script to scrape LinkedIn invitations is below. You can only see 100 invitations per page, though.

copy(
  $$(".invitation-card").map(v => ({
    name: (v.querySelector(".invitation-card__title") || {}).innerText || "",
    link: v.querySelector(".invitation-card__link").href,
    subtitle: (v.querySelector(".invitation-card__subtitle") || {}).innerText
      || "",
    common: (v.querySelector(".member-insights__count") || {}).innerText || "",
    message:
      (v.querySelector(".invitation-card__custom-message") || {}).innerText
      || "",
  })),
);

PS: A similar script to scrape LinkedIn people search results is below.

copy(
  $$(".entity-result").map(v => {
    const name = v.querySelector(
      ".entity-result__title-text [aria-hidden=\"true\"]",
    );
    const link = v.querySelector("a");
    const badge = v.querySelector(
      ".entity-result__badge [aria-hidden=\"true\"]",
    );
    const title = v.querySelector(".entity-result__primary-subtitle");
    const subtitle = v.querySelector(".entity-result__secondary-subtitle");
    const summary = v.querySelector(".entity-result__summary--2-lines");
    const insight = v.querySelector(".entity-result__simple-insight-text");
    return {
      name: name?.innerText || "",
      link: (link?.href || "").split("?")[0],
      badge: badge?.innerText || "",
      title: title?.innerText || "",
      subtitle: subtitle?.innerText || "",
      summary: summary?.innerText || "",
      insight: insight?.innerText || "",
    };
  }),
);

Comments

mvark 8 Aug 2021 10:39 pm: I think rather than create JSON, building a comma separated string and console.log-ging that string will show the whole list in the Dev Tools Console. Using the Save As option in the context menu, the list can be saved as a .CSV file directly without Step 3 & 4.
Tools to publish annotated talks from videos - S Anand 20 Oct 2024 12:53 pm (pingback): […] together. As you scroll, newer/older comments are loaded. So I needed to use my favorite technique: Cyborg Scraping. During Q&A, I kept scrolling to the bottom and […]

email analysis

Wed, 16 Feb 2005 12:00:00 +0000

Marc Eisenstadt has analysed 15 years of email.

… it is trivially easy to get to 2.5 hours per workday assuming a fairly ruthless, ‘one-touch’, knee-jerk email interaction regime. And worse if you deviate from the regime.

Then there are other sources of workflow: blogs, aggregator summaries, phone calls (rare, but I still allow one or two), cell-phone, text message, instant messaging (my buddy list is very large, and most of them are work-related).

Interesting that Knuth opted out of email in 1990.

Comments

Aditya Chaturvedi 16 Feb 2005 12:00 pm: Interesting Analysis
m1108739289513 16 Feb 2005 12:00 pm: See this dilbert on email and its sequel next day
S Anand 16 Feb 2005 12:00 pm: After reading this article, I turned off my e-mail alert. It is amazing how much distraction it reduces.