Timely Tome: How I Built a Literary Clock

"Time flies over us, but leaves its shadow behind." - Nathaniel Hawthorne

I've always loved the idea of the Author Clock, a clock that tells time through literature. Instead of numbers, it shows a quote from a book that mentions the current time.

So I built my own. I call it the Timely Tome.

It sounds simple on paper: find quotes that mention times, display them on a screen. But it turns out that teaching a computer to understand time in prose, dealing with wildly uneven data, and displaying it all on hardware that refreshes like a Kindle... that's where things get interesting.

Here's how it works.

---

Finding Time in Text

The first problem is deceptively hard. I needed to scan thousands of books and pull out sentences that reference a specific time of day.

You can't just Ctrl+F for "3:00." A sentence like "I waited for three hours" is a duration, not a time. "The clock struck four" , is that 4 AM or 4 PM? And "it was nearly dawn" doesn't mention a number at all, but it clearly implies a time.

I built a pipeline in Python using NLTK to process raw text from Project Gutenberg and other archives. Try the interactive visualizer below to see how the system "reads" a sentence:

How the pipeline works

1. Raw Input: The system ingests raw text. Punctuation and formatting are messy.
2. Entity Extraction: Regex and Named Entity Recognition (NER) tag potential time phrases.
3. Context Window: A quote like "It was 4:17" is boring on its own. I used an LLM to expand the selection to surrounding sentences , enough to give the quote flavor and narrative context, without making it too long to fit on a small screen.

---

The Distribution Problem

This is where I hit a wall I didn't expect.

After scanning over 2,000 books, the data was wildly uneven. Authors don't write about every minute of the day equally , not even close.

• 4:00 PM and 8:00 AM are mentioned constantly (end of school, start of the workday).
• Midnight is a favorite for dramatic scenes.
• 4:17 AM? Almost never.

I visualized the number of quotes I found for each hour. Hover over the chart to explore:

Filling the gaps

The valleys in that chart were a real problem. If it's 3:33 AM and I have zero quotes, the clock just... breaks.

My solution: fuzzy matching. If the exact minute isn't available, the system looks for "fuzzy" time descriptions , things like "It was just past three-thirty" or "The middle of the night." I assign these to the empty slots to ensure 100% coverage across all 1,440 minutes, even if some are less precise than others.

---

The Hardware Reality (e-ink)

Finally, displaying the text. I chose an e-ink display (like a Kindle) because it emits no light and looks like real paper , perfect for sitting on a bookshelf.

But e-ink is physically strange compared to an LCD. It uses tiny capsules of black and white pigment, moved around by electric charges. This means it can't update instantly.

Why does it flash?

If you've ever used an older Kindle, you've seen the screen flicker black before turning a page. That's a "full refresh" , it's clearing ghost images, the faint outlines left behind by previous text.

• Ghosting: Without a full refresh, old text bleeds through. It looks like a palimpsest (which is kind of poetic for a literary clock, but not great for readability).
• Temperature: In cold rooms, the pigment moves slower, so refresh cycles take longer.

I wrote a custom Python driver that manages all of this. It does a "quick update" (no flash) for most minutes, but forces a full refresh every 15 minutes to keep the screen crisp.

---

Architecture Overview

For the curious, here's the full stack that runs the clock:

• Frontend: Next.js (the web demo you're reading right now).
• Backend: Python scripts running on a Raspberry Pi Zero W.
• Database: A simple JSON store of processed quotes , fast lookups, no SQL overhead needed.
• Deployment: A systemd service on the Pi launches the clock on boot.

What I Learned

Building Timely Tome was an exercise in trade-offs. Precise NLP vs. processing time. Perfect data distribution vs. the reality of what authors actually write about. Beautiful hardware vs. painfully slow refresh rates.

Every engineering decision involved giving something up. But that's what makes the final product, a quiet, literary clock on my shelf, feel all the more alive.