About
Eryawan Presma Yulianrifat
I build production AI systems while researching how models work over documents.
I started in programming contests, moved into language-model research, and now work as a founding engineer at Klaussa, where we apply those systems to Indonesian regulation.
- Klaussa
- Co-founder, founding engineer
- Seed-funded legal technology, in production
- IRNLP Lab, Universitas Indonesia
- Research intern
- Information retrieval and language, under Dr. Alfan Farizki Wicaksono
- EleutherAI
- Mentee
- Summer of Open AI Research, a mentorship programme
- Indonesia–UChicago SYAIR
- Programme member
- A research programme led by Prof. Haryadi Gunawi, University of Chicago
The path
Contests, then research, then a company.
2021 to 2023
Contests first
Competitive programming came first. A silver medal at the Indonesian National Olympiad in Informatics carried direct admission to Universitas Indonesia with no entrance exam, and two runs at the Asia Jakarta regional of the International Collegiate Programming Contest finished at rank 15 with an honourable mention.
I stopped competing in my second undergraduate year and put the time into research and building products.
Source · Competition placements, each listed with the size of the field it was won against
Since 2024
Research
His research is in natural language processing, agentic large language models and reinforcement learning. Those three meet whenever a model has to carry out work over a body of documents.
Alongside building at Klaussa, I research information retrieval, agentic language models, and reinforcement learning through work with IRNLP Lab, EleutherAI, and Indonesia–UChicago SYAIR.
Source · His own account of the positions he currently holds
Since 2024
Klaussa
He co-founded Klaussa and is its founding engineer. Klaussa is a seed-funded legal-technology company whose product applies large language models to the body of Indonesian regulation.
His work covers the whole system: collecting the documents and running the in-house OCR that turns them into text, building the citation graph that links a regulation to the ones it amends, training the Klaussa-8B model, and running the agent and the Next.js and FastAPI application that serve it.
Source · Klaussa's own corpus and model figures
Next
What the work is for
I want research that survives contact with production, and production systems with enough method underneath them to be worth trusting.
Source · His own stated aim
The research record
Published during undergraduate study.
8papers
40citations
Natural language processing, agentic LLMs, reinforcement learning.
Both figures are the ones reported by his Google Scholar profile, which is public and can be checked.
Selected · 4 of 8, by venue
- SemEval 2025How well current text encoders detect several emotions at onceFirst author
- ACL 2025SEA-VL, an image and text dataset covering Southeast Asian culturesCo-author
- IWSLT 2025Speech translation from Maltese to English, a language with little training dataCo-author
- Daengku 2024Gender disparities in Indonesia's National Science Olympiad resultsCo-author
These are the papers whose venue and authorship position are listed. The rest of them, and the citations they have drawn, are on the Google Scholar profile.
On the record
Placements, with the size of each field.
Each line gives the number of teams or candidates the placement was won against, because a rank means little on its own. Placements whose field size he cannot state are left off.
Programmes and selection
Competition
He last competed in his second undergraduate year.
If you are a student, researcher, or founder working on something adjacent, send it over. I’ll tell you what I would investigate first.