How rafaelbastos44/tcc works
Project type and purpose
• This is a research‐oriented text‐classification pipeline, not a web or full‐stack application.
• Purpose: compare zero‐shot LLM approaches (both API and local) versus supervised methods (scikit‐learn, neural nets) across binary, multiclass, and multilabel classification tasks, measuring performance, time, and cost.
File structure
• Top level: README.md, requirements.txt, .gitignore, and a single “text_classification” folder.
• text_classification/data: raw and preprocessed datasets.
• text_classification/LLMs: zero‐shot pipelines subdivided into api_llms (openai, gemini) and local_llms (deepseek, gemma, llama).
• text_classification/neural_nets: BERT grid-search & training, LSTM, modernBERT experiments.
• text_classification/sklearn_models_revised: scikit-learn experiments, models, and results.
• text_classification/analysis & comparison.ipynb: final comparative analysis across methods.
README for additional insights
• Confirms the goal: build a well‐structured pipeline to identify best classification approach for different contexts.
• Data from open platforms; metrics include performance, latency, and cost.
Elements
• Clear labels: “Data Ingestion,” “API LLM Pipeline,” “Local LLM Pipeline,” etc.
– Rectangles for compute modules (pipelines)