Akim Mousterou | ムステロ・アキム

AI/NLP Engineer from Paris, France | パリ、フランスからの自然言語処理エンジニア

• Master's degree, in Natural Language Processing - Multilingual engineering, Japanese at I.N.A.L.C.O

(I.N.A.L.C.O 東洋言語文化学院、自然言語処理修士 | 卒業)

• Master's Degree in International Business (EMIB) at ESCP Europe

(ESCPヨーロッパ・ビジネススクール、経営管理修士 | 卒業)

I have managed AI and business initiatives for prestigious brands, publishing companies, technology companies, and financial institutions for over 10 years. Born and raised in Paris, I am fluent in French, English, Japanese (JLPT N2), and Spanish. I am passionate about quantitative finance, network effects, opinion mining, and natural language processing.

Research papers:

• オープン日本語LLM リーダーボードの構築と評価結果の分析

  ("Building Open Japanese LLM Leaderboard and in-depth analysis of primary results" for ANLP 2025 - March 2025 in Nagasaki)

• NER-Luxury: Named entity recognition for the fashion and luxury domain

  (arXiv:2409.15804 - Submitted on 24 Sep 2024)

• LLM-jp: A Cross-organizational Project for the Research and Development of Fully Open Japanese LLMs

  (arXiv:2407.03963 - Submitted on 4 Jul 2024)

Open-source contributor to LLM-jp in the "evaluation" working group led by Prof. Yusuke Miyao (University of Tokyo - 東京大学).

Please feel free to connect with me on GitHub, LinkedIn, or HuggingFace ! 😃

---

🇪🇺 Open-source RAG Mixtral 8x7B with financial reports in German, French, and Italian

Libérté, égalité, architecture open-sourcé... Open-source RAG LlamaIndex and multilingual LLM from Mistral AI in a low-resource environment over financial statements:

• Vanilla RAG (retrieval-augmented generation) with LlamaIndex and pgvecto.rs from TensorChord
• Quantized model of Mistal8x7B from Mistral AI and LlamaCPP inference tool
• Test of BGE-M3 embedding model from BAAI
• Plus benchmark in German on Porsche AG, in French on Hermès, and in Italian on Brunello Cucinelli

🎆 Open-source RAG for Japanese LLM in low-resource settings

Generative AI for all - Quick implementation with an open-source RAG LlamaIndex and Japanese LLM from ELYZA, Inc. in a low-resource environment over legal documents:

• RAG (retrieval-augmented generation) is LlamaIndex with a vanilla Hybrid search (combining retrieval from both text search and vector search)
• Japanese LLM “ELYZA-japanese-Llama-2-7b-instruct” created by Japanese startup, ELYZA, Inc.
• Open-source database PostgreSQL transformed into a vector database by the great library PG Vector
• Plus Q&A analysis in Japanese, embedding pricing war, and generative AI strategy of France, USA, and Japan

💎 A fine-tuned XLM-Roberta model for NER in the fashion and luxury industry

NER-Luxury is a fine-tuned XLM-Roberta model for the subtask N.E.R (Named Entity Recognition) in English. NER-Luxury is domain-specific for the fashion and luxury industry with bespoke labels. NER-Luxury is trying to be a bridge between the aesthetic side and the quantitative side of the fashion and luxury industry.

• 38.063 sentences in English (covering the beauty, fashion, and luxury industries)
• 32 labels from companies, groups, and holdings to luxury brands, models, and magazines
• Loss: 0.3990, Accuracy: 0.9427, F1: 0.7879

⚕️ English, Japanese, and French Anki flashcards about Natural Language Processing

To promote communication between NLP practitioners, I created a vocabulary deck of 420 technical words for Anki Pro (learning software based on space repetition) in Japanese, English, and French for natural language processing.

🎼 Inference testing on earning calls with "Whisper" by Open AI for Japanese

• Test on earning calls of Fast Retailing Co., Ltd 2022 with Whisper OpenAI

• LDA analysis on shareholder's letter

• Sustainability as a strategy in fashion and in NLP

  

🐈‍⬛ Preliminary research about Quantum Natural Language Processing for Japanese

• Introduction from Pregroup Grammar, DisCoCat, to Lambeq
• Specificities of the Japanese language
• Pregroup Grammar in free word order

📊 Quantative analysis of Ferragamo - Black Scholes & 🇪🇺 Options in Python & C++

• Quick financial analysis
• Why is TikT0k a mistake in luxury?
• Black Scholes (closed form) and the Greeks implemented in Python and in C++

⚾ Named-entity recognition in Japanese with Masayoshi Son (Softbank) on Twitter

• Linguistic specificities for NLP in Japanese
• Leveraging spaCy framework, and Ginza model, and building a custom NER model
• Insights on Masayoshi Son (Softbank), Hiroshi Mikitani (Rakuten), and Haruhiko Kuroda (Bank of Japan)
• Few thoughts on NLP in Japanese

🇮🇹 Forecasting analysis of Moncler s.P.a with Prophet and Bayesian inference

• Naive OLS Linear regression model
• Facebook's Prophet model
• Bayesian Linear regression model with STAN
• And a quick introduction to the Hierarchical model

🇺🇸 Meme strategy - Polarity analysis of Elon Musk on Twitter from Apr to Dec 2020

• Natural Language Processing using NLTK and Vador
• Time Series for sentiment analysis
• "Meme" as a marketing or investment strategy?
• Insights on the human perceptions of leadership on Twitter

🩺 Covid-19 - Financial performance of listed luxury groups

• Financial analysis of luxury groups in 2020 during a high volatility state with the spread of SARS-Cov2
• Found an increasing correlation on Y-o-Y between luxury groups and the Facebook Inc. ecosystem
• Conclusion for stakeholders and shareholders