Welcome to Paper Lanterns
Why I love letters, how this started alongside Project Akshara, and why everything here is free.
By Bhuvan
Six minutes

Hello, fellow traveler. Welcome to Paper Lanterns.
This is my passion project, born out of discovering the absolute wonderfulness of historical correspondence and falling in love with it.
The inspiration for Paper Lanterns comes from Shaun Usher’s Letters of Note. It was a newsletter I accidentally discovered, and I absolutely loved reading all the letters Shaun was curating and publishing. Some of these letters were really heartwarming, funny, and beautiful. It was a delight to read them. I wanted to do something similar, while also focusing on some of the more unknown letters.
Why letters?
You see, letters are unique because most letters were never meant to be read by a large audience. They were often intimate messages, meant to be read by at best one person, with a few exceptions.
The thing I’ve noticed about letters is that the way an author writes a letter is not the same way he or she writes a novel or a book. Take Dostoevsky’s letter to his brother from prison, which is one of my favorite letters. It’s just absolutely brilliant:
Brother! I have not become downhearted or low-spirited. Life is everywhere life, life in ourselves, not in what is outside us. There will be people near me, and to be a man among people and remain a man for ever, not to be downhearted nor to fall in whatever misfortunes may befall me—this is life; this is the task of life.
Even though I’ve not read Dostoevsky’s books, I’m reasonably sure that Dostoevsky the writer is different from Dostoevsky the brother. Take Hemingway’s case too. In a lot of cases, that’s what makes letters truly unique: they showcase a different side of some of the greatest people in history—the men and women we admire, whose books and poems we read.
Oh, as an aside, you can now read the great Emily Dickinson’s letters here too. Her verse is one of my all-time favorites:
I dwell in Possibility —
A fairer House than Prose —
More numerous of Windows —
Superior — for Doors —
Anyway, getting back to the point: that’s why I love letters. They often showcase a whole different side—mostly good, loving, wonderful, tender, warm, affectionate, and caring, and in a lot of cases, the not-so-good sides of humans too.
And they also often have some, to use very modern lingo, bangers. For example, in his letter to his uncle, Lord Burghley, Francis Bacon uses the line:
for I have taken all knowledge to be my province.
That line is somewhat of a philosophy that I try to live by. I have not yet digitized Francis Bacon’s letters as a collection, but it’s on my to-do.
The backstory
Not that you asked, but let me share a brief backstory of how this came to be.
The history of this project dovetails with another project I was working on: Project Akshara (I wrote more about it in my latest post on the Akshara blog). Both of them started around the same time.
This was also when LLM coding tools were starting to get better, so I set up a simple website and started curating some of the most famous letters in history. I had about 40–50 letters on the site.
Then the project took a backseat because another question was bothering me. Despite all our boasts about being the oldest civilization—India gave this to the world, India gave that to the world, Indian history is this, Indian history is that—why wasn’t there a Project Gutenberg for India? Why had we done such a piss-poor job of preserving our written memory?
I wanted to do something about this, but I was hamstrung by my lack of technical skills, so nothing much came out of it. Then, as coding tools were getting better, vision-language models were getting better too—often much, much better than traditional OCR tools. As I was hacking and messing about with these models, they got good enough for me to start digitizing books. I figured I might as well do it in parallel for letters, so I spun out a new subsection on Paper Lanterns called Collections, where I started uploading whole collections of letters.
Again, it took a backseat because I was trying to do the same manual checks on letter collections that I was doing with books on Akshara: front matter, table of contents, heading schema, formatting, diacritics, scripts from other languages, footnotes and whether they were properly linked, tables, numbers, margin notes, archival fidelity. All of this used to take an entire day for one book, which simply wasn’t scalable. And while checking books on both Akshara and Paper Lanterns, I physically started hurting from staring at screens for that long. It wasn’t sustainable, and it was clearly unscalable.
What I decided to do with both projects was build a community around them—see if there are other people interested in preserving these letters and making them readable, and get people involved in proofing and checking these books alongside the vision-language models. (You can read more about how the pipeline works on How these letters are read.)
The LLMs now do the bulk of the heavy lifting. Accuracy is well north of 90% on pretty much all letter collections—in fact, I’m sure the extracted text is well north of 99% in most cases. There are very few errors, mostly specific to a particular collection or formatting. That’s the price you pay for relying on large language models; maybe they’ll get better.
You can still push the accuracy higher, of course, but that requires frontier models, and when it comes to digitization tasks, frontier models are ridiculously expensive. I pay for all of this from my own pocket—I don’t have any benefactors or patrons—so I rely on relatively cheaper models, mostly the Flash family with Gemini, and in some cases Sonnet because I get some free API credits with my Claude plan. They’re cheaper, but they’re not bad; they’re really good at maintaining archival fidelity, with just a few issues here and there.
All of this is possible only because the Gemini models are so good—both at creating the pipeline and at digitization—and cheap enough for me to do this by setting aside some money from my monthly salary. I would love to do this at scale, but the costs add up. So if anybody from Gemini is reading this, like I asked on Project Akshara too: if you give me some free credits, I promise I’ll put them to good use, digitizing historical collections of letters and putting them back into the public domain.
Sorry for the long, rambling post, but that’s the story behind Paper Lanterns.
Please enjoy the letters here. Everything is free, everything is in the public domain. You can use it and reuse it however you see fit. All I ask is a little credit to the site, nothing more. No money, no nothing. I have nothing to sell.
Again, people from Google, if you’re reading—please, would love some credits. 😬
And the other help I need: if you know this area of letters and correspondence, if you’re an expert or just interested in it, and there are any letter collections worth digitizing that I should know about, please reach out to me. Or if you have any suggestions, advice, brickbats, kudos, ideas, whatever it is—just reach out.
You can email me at bhuvan@bebhuvan.com, or find me on Twitter (@bebhuvan), LinkedIn, or Substack.