Show HN: I am building a map of people who lived in the Roman Empire

4 days ago (new.roman-names.com)

Driving home from work one day, I wanted to know how many people we knew the names of who lived during the Roman era. Searching around, I found lists of Consuls and officials, but nothing that covered ordinary people or even most people like freedmen and slaves. So I ended up building a pipeline to process the more than 500k Latin inscriptions in the Epigraphic Database Clauss-Slaby https://edcs.hist.uzh.ch/en/ and extract the names of people (and attempt to cluster them, but this is a work in progress).

There are databases where Classicists have done this manually for specific regions, Trismegistos https://www.trismegistos.org/ and Latin Inscriptions of the Roman Empire (LIRE) https://pure.au.dk/portal/en/publications/latin-inscriptions... are two major efforts I found. But there doesn't seem to be a project that did what I set out to do, although I have read in some places that it was believed to be possible.

I am not a classicist or a web developer, but I have Claude and Gemini and I can sort of read basic Latin - so I set to work. I used LIRE and another database as ground truth and built a pipeline to extract and process the inscriptions to recover the names. The process I developed uses a high end LLM like Sonnet or Gemini Pro to supervise the extraction and tuning process on a regional basis until the obvious error rate is reasonable. For this, so far, reasonable to me means less than 1-2% in the smaller initial samples of 100-500 and no observed systemic issues. The different regions often need different prompts, so this basically became an exercise in letting the higher level AI tune the prompt for the lower level AI. The extraction when measured against LIRE produces an F1 score between 0.64 and 0.87, but take this with a grain of salt.

Once I had done a few regions, I wanted to see the work, so I threw together a pretty crude website but as I am not a web developer, it was crude in how it accessed its data. It does look cool and I also added summarization, and machine translation to each entry. I wanted to eventually get feedback from an actual team of classicists and make the website work better, so I am rewriting it as we speak but it is broadly functional now with a few extra bugs but substantially improved performance compared to the old one. All entries link back to the proper sources, and the old web app linked to several additional sources where the data was present, but I haven't gotten that working again just yet on the new one. (The old web interface is still available at https://roman-names.com, but I will warn you it is clunky and not mobile friendly at all)

Key findings so far:

AI supervised AI extraction saved me time. I was manually tuning things for a while and then the runbook became an idea that I feed my instructions in and let the big AI go with sparse oversight from me.

The extraction improved significantly (by about 10 F1 points) when I fed the model the raw text including the markers, vs a cleaned up version of the text.

I just thought it was a cool little project and wanted to share. If you happen to work in any adjacent space and there is something I could do better etc let me know.

48 comments

metiscus

Reply
jnovek  20 hours ago

I guess before I roll out questions and criticisms, I just want to say that this is a really cool project. I love it.

Could you make the dots smaller in the updated UI? I didn’t realize at first that you were using an actual map of Roman provinces.

My eyesight isn’t great and it would help if you used a political map rather than terrain. I’m not sure what’s out there for ancient Roman map tiles, though.

I’m not so much of an antiquity scholar AND I’m an American so my European geography isn’t perfect. It would be neat to be able to flip to a modern map, too, so I can see where things are in terms of modern landmarks.

You’re not getting a ton of comments so far, but FWIW these are the kinds of projects I come to HN for. I’ve been getting into opera lately and suddenly classical antiquity is very relevant to my interests. I’m going to keep this in my bookmarks, I’m finding the tangential historical stuff related to opera is drawing me in nearly as much as the music.

I’m also going to pass it on to an academic friend of mine who is working in an unrelated field but might find similar techniques useful.

Finally, when I first opened the map, I recognized the basic shape of the peak Roman Empire in the dots! I love when data does that kind of thing.

Thank you again for sharing this very cool project.

1e1a  1 hour ago

This is very cool, but unfortunately it's hard to get a sense of what the actual spatial distribution of people looks like when nearby nodes are grouped into larger dots that obscure the local structure.

cwnyth  19 hours ago

Hi, I'm an actual classicist (phd and all), if you wanted to throw any questions my way.

  • metiscus  12 hours ago

    I've contacted a Professor in Europe who was doing research in this area and pointed him to the page. What I genuinely need is someone to spot check a few of the attributions. I can send you a list of the ones I think are the most likely to be good.

      1. Laepoca / Laepocus — Piquentum, Venetia et Histria (1–50 AD)
  Three family members: two women (Laepoca Regilia, Laepoca Tuia) and a man (Metellus Laepocus). The nomen appears in both feminine and masculine forms in the same inscription, pointing to a
  genuine local gentilicium, likely of Istrian or Liburnian origin.
  https://new.roman-names.com/#edcs_id=EDCS-04200530

    It looks like my auto-translation and summarization layer is hallucinating on this entry, but the extraction appears correct. I'll flag it for the next run.

      2. Tocernius — Eraclea Veneta, Venetia et Histria (3rd c. AD)
  Father (C. Tocernius Hermeros) and son (C. Tocernius Maximianus), the latter a soldier of Legio II Italica. Probably a Venetic name surviving into the imperial period.
  https://new.roman-names.com/#edcs_id=EDCS-04200461

    Here, the auto-translate and summary worked as intended. It does garble the dedication into the status.

      3. Laulenia — Thibilis, Numidia
  Two sisters, Laulenia Matrona and Laulenia Naxina, daughters of the same Marcus. The name looks Berber/Numidian in origin. (I should note that our pipeline transcribed the nomen as
  Lauzenia — the raw EDCS text reads Laulenia, which is probably the correct form.)
  https://new.roman-names.com/#edcs_id=EDCS-13500401

    The auto-translate and summary layers do not make this error, only the name extraction layer does. I have flagged the entry and am diagnosing it.

      4. Kanulanius / Nansinia — Flavia Solva, Noricum
  Father (C. Kanulanius Eumitus) and son (C. Kanulanius Nepos, a soldier of Ala III Thracum). The K-spelling may reflect local Celtic orthographic convention. The wife's nomen, Nansinia,
  also appears unattested in standard sources and may be a second find in the same inscription.
  https://new.roman-names.com/#edcs_id=EDCS-14500644

    Here there is an issue where I think in the processing for the web I am feeding interpreted text into the raw extraction field as my displayed raw text seems to be expanded from EDCS. Mine: Caius Kanulanius Eumitus vivus fecit sibi et Nansiniae Verecundae coniugi et Caio Kanulanio Nepoti filio militi alae III Thracum annorum XXV stipendiorum VI loco et impensa Anni Festi

    EDCS:

    C(aius) Kanulani/us Eumitus / v(ivus) f(ecit) sibi et / Nansiniae / Verecundae con(iugi) / et C(aio) Kanulanio / Nepoti f(ilio) mil(iti) alae III / Thrac(um) an(norum) XXV stip(endiorum) VI / loco et impensa / Anni Festi

    • cwnyth  6 hours ago

      What exactly are you trying to go for with status? It seems it mostly records filiation, but I don't think that's an intuitive use of the word. Knowing what you're actually going for would be helpful.

      Also, you might want to include the source from EDCS. #3 above comes from ILAlg, and EDCS has a key for all the collections and their abbreviations. This will help someone be able to track down the original inscription more easily.

      1. That first one is rough, and the translation is broken (it doesn't even translate Surus' name), but you got the people down. Regilia is just a guess, though.

      3. Yep, Laulenia is the original name. Seems like AI is hallucinating here.

      4. Have you thought about code that strips the parenthesis first, instead of letting AI do it? Also, loco et impensa is something like "grave site and expense", not "expense and initiation." Locus means "place", and in epitaphs often just refer to the burial place.

busyant  12 hours ago

Very cool. I'd like to explore it, but I'm getting a lot of "page unresponsive" and I can't tell if it's a back-end server being swamped, or if it's something else.

My son just graduated with a double major in classics and molecular biology (doing informatics work), so maybe he could help! lol

goldfishgold  7 hours ago

On the Greek side this has been done, the Greek Lexicon of Personal Names is an effort to find all Greek names down to 600 CE. https://www.lgpn.ox.ac.uk/. You might want to try using that as the basis of your map.

In general the word 'prosopography' might be helpful for you. There's been lots of work over the centuries on analyzing large groups of people in antiquity.

trevoragilbert  20 hours ago

This is very cool! For the name extraction, how are you handling false positives across such a large dataset? I’m assuming there are mentions that could be a name but are actually just a noun. For example, Agricola being the word for farmer but also a name.

  • metiscus  12 hours ago

    So most inscriptions are somewhat formulary, and I provide examples to the llm to assist it to find the names. I also have a postprocess blacklist that removes some known cases where things slip through. It's never going to be 100% perfect but to my untrained eye, it seems to do okayish. Waiting on some professionals to cross check my data. If that is you, you can search and export the data in csv via the browse button.

doodlesdev  11 hours ago

Since you are using LLMs to create the transcriptions, I wonder whether you've measured the difference in precision between the chosen model, Gemini 2.5 Flash-Lite, and newer/larger models such as Gemini 3.5 Flash, Gemini 3.1 Flash-Lite or GPT5.5.

I've read the README in the feat-api branch and, from what I understand, you've already assessed that false negatives are not a model failure, but I'm not sure I understand why (haven't spent that much time looking at it though, just curious to hear from you).

This is a really cool project, by the way! In my opinion this is a place where LLMs shine: produce the work of hundreds of hours of manual human labor much quicker and cheaper, for something that no one else would ever bother to do the work!

yubblegum  19 hours ago

> I have Claude

And just now I am watching I, Claudius.

aspenmartin  19 hours ago

This is really wonderful -- One thing that may be really cool if you have the data is to add a time-axis ability (unless I missed it) for a given location. This is such a delightful application of AI!

  • metiscus  12 hours ago

    Yeah, I thought about that. The issue is that the date density is poor already and the ranges are pretty broad. Someday I'll give it a try, but the search interface can limit you to date ranges for now, so the infrastructure is all there.

daviTeodoro  20 hours ago

Very cool! Do you plan to share the final dataset? I've been working with geographic data all my life and I'm building a Carto/Felt alternative. Do you want to have your data their? It is https://cartografo.io/ There is a price tag, but I can host this dataset for free for you. I would love to have this map there to show case. If you are interested send an email, davi@cartografo.io. If you just need some help improving your map I can help you as well.

metiscus  3 days ago

I am hoping to push a few fixes to the new web interface later today, so if you looked at this and saw anything off, hopefully by COB today I will have the known issues fixed.

  • metiscus  11 hours ago

    And thank you to whomever flagged EDCS-31400647 and EDCS-21700214. The militia militia militia attribution was hilarious, wrong, and has been fixed.

OJFord  15 hours ago

This is cool – I'd like a way to click through from a given entry to find out more, appreciate it may not be much, but each one came from somewhere. There's a few in my hometown so just idly interested if they're from somewhere I know about or something that would be new to me.

  • metiscus  12 hours ago

    so there should be some links when I have the information available. If you link me the entries I'll see what is going on.

frereubu  19 hours ago

This is great. One little bit of UI feedback: the green map clusters when zoomed in quite a bit aren't very obvious on green backgrounds - they merge into the background features a bit (e.g. in the very west of Scotland).

bvan  2 hours ago

This is the kind of application which brings me back to shore as I feel I’m drowning in AI slop. Really well done and thanks for sharing.

ingvay7  19 hours ago

Love this. For people who aren’t familiar with Roman history, it would be great to have a short guided tour of how to explore the map. I filtered for 'pompeii' and it gave me 117 dots.

avyeed_desa  19 hours ago

Congrats to this great idea! Love it.

The ones around my place all use EDH, which also has a map feature, but not as intuitive as this! Reminds me of vici.org

  • metiscus  12 hours ago

    I also used EDH as ground truth but not directly as a source and link out to it and other sources when I have those links available.

andai  20 hours ago

Could you elaborate on the multi level LLM workflow? Did you set up a benchmark, and you're having a LLM mutate prompts?

  • metiscus  12 hours ago

    But yes, the runbook in the project gives the llm instructions on how to use the scripts and what to modify. I let Claude code read that and tell it to work on a province. It runs a small segment and analyzes the results until it hits 1-2% error with no systemic errors. If it can't get all the errors out then I have it switch to using gemini-flash-lite-latest instead of 2.5, which costs slightly more but performs much better. Basically Claude code runs a self governing loop with my oversight mutating the prompts and data inputs to extract all the names.

    EDIT My instructions to the supervising LLM are in here https://github.com/metiscus/roman-names/blob/feature/webapp-...

oezi  20 hours ago

Very nice. Are you using the roman roads which were also on HN a couple of weeks ago?

jdthedisciple  20 hours ago

Very nice idea but please, check the performance: I had to close the tab 3 seconds in because it got stuck and my CPU fan got noisy.

So I couldn't even check it out properly.

tonymet  17 hours ago

Great resource for baby names. And a great way to help students better understand historiography.