Artificial Intelligence and Machine Learning

Computational methods
with replicable method.

Five services covering machine learning, natural language processing, computer vision, and generative AI applied to research questions. Operations run on PyTorch, scikit-learn, HuggingFace, OpenCV and related tools, with output ready to defend in peer review — metrics, interpretability, and documented ethical protocol.

Predictive Modeling

When the goal is to predict, not just explain — and the model must sustain that promise.

Classification, regression, clustering. Feature selection with statistical and model-based methods, hyperparameter tuning via grid search or Bayesian optimization, stratified cross-validation, problem-appropriate metrics (AUC-ROC, F1, RMSE, MAE), interpretability via SHAP, LIME or permutation importance. Output includes a technical report ready for the results section, comparative tables of tested models, and publication-quality figures. Documented code available as add-on.

NLP and Text Mining

Text is also data — when properly processed, it reveals patterns that close reading cannot reach.

Sentiment analysis, topic modeling (LDA, BERTopic, Top2Vec), document classification with transformers or classical models, named entity extraction, corpus analysis with lexical measures. Applicable to social networks, electronic medical records, legislative documents, jurisprudence, digitized historical archives, transcripts. Pipeline covers corpus-appropriate preprocessing (tokenization, lemmatization, OCR noise treatment when applicable), embeddings via pretrained or custom models, and result validation with human-in-the-loop when the domain requires.

Computer Vision

When the research question lives in thousands of images — and manual annotation does not scale.

Image classification, object detection, semantic segmentation, video anomaly analysis. Fine-tuning of pretrained models (CLIP, YOLO, ResNet, ViT) for specific domains with transfer learning when the annotated set is limited. Applicable to medicine (radiography, histopathology), biology (microscopy, ecology), civil engineering (structural inspection), remote sensing, visual studies in art and culture. Pipeline includes image preprocessing, augmentation, stratified cross-validation, and appropriate metrics (mAP, IoU, dice score).

Generative AI Applied to Research

Generative AI is an analytical tool, not a ghostwriter — and using it with integrity requires protocol.

Use of language models (GPT, Claude, Gemini, open models) for analytical tasks: automated classification of open responses, large-scale literature summarization, initial screening for systematic review, categorization of extensive corpus, evidence extraction from documents. Aria does not replace the researcher — it integrates AI as a tool with documented ethical protocol, sample-based validation, and full transparency about use in the final manuscript per editorial guidelines (COPE, ICMJE).

Complete Data Science Pipeline

When the project requires the full operation — from collection to interpretation — without changing providers along the way.

End-to-end: collection, preprocessing, feature engineering, modeling combining classical statistical and computational methods, rigorous evaluation, interpretability, methodology and results section writing ready for submission. For projects where the research question requires technical integration that isolated analyses do not deliver — combination of classical quantitative methods with machine learning, triple validation (statistical, computational, substantive), and coherent methodological narrative. Documented code included by default.