richi ibnu wardana

About

Built an automated ML/NLP document classification system for unstructured legal documents, combining classical models (TF-IDF + logistic regression/random forest) with entity resolution via fuzzy matching validated by precision/recall. Also implemented semantic similarity search using sentence embeddings stored in FAISS and improved matching by fine-tuning a transformer on domain-specific data and tuning similarity thresholds for fewer false positives.

Key Strengths

• Built ML/NLP document classification for unstructured legal text
• Strong text feature engineering and classical ML modeling (TF-IDF + logistic regression/random forest)
• Entity resolution using fuzzy matching with precision/recall validation and manual review
• Implemented semantic search/entity linking with sentence embeddings and FAISS
• Fine-tuned transformer/embeddings on domain data and tuned similarity thresholds to reduce false positives
• Production workflow practices: modular Python, Airflow scheduling, tests, documentation
• Balances experimentation with delivery using accuracy/response-time metrics and user feedback