llmclean (Q7376804)

Detects and suggests fixes for semantic inconsistencies in data frames by calling large language models (LLMs) through a unified, provider-agnostic interface. Supported providers include 'OpenAI' ('GPT-4o', 'GPT-4o-mini') <https://platform.openai.com>, 'Anthropic' ('Claude') <https://www.anthropic.com>, 'Google' ('Gemini') <https://ai.google.dev>, 'Groq' (free-tier 'LLaMA' and 'Mixtral') <https://groq.com>, and local 'Ollama' models <https://ollama.com>. The package identifies issues that rule-based tools cannot detect: abbreviation variants, typographic errors, case inconsistencies, and malformed values. Results are returned as tidy data frames with column, row index, detected value, issue type, suggested fix, and confidence score. An offline fallback using statistical and fuzzy-matching methods is provided for use without any application programming interface (API) key. Interactive fix application with human review is supported via 'apply_fixes()'. Methods follow de Jonge and van der Loo (2013) <https://cran.r-project.org/doc/contrib/de_Jonge+van_der_Loo-Introduction_to_data_cleaning_with_R.pdf> and Chaudhuri et al. (2003) <doi:10.1145/872757.872796>.