Subtext

neurodivergent people struggle to understand tone, sarcasm, and intent in messages leads to miscommunication, anxiety, suboptimal responses

This project essentially creates a platform through which neurodivergent users can input their text conversations in order to understand tone, acronyms, and even native context

We built it through: Stack: Python, Streamlit, scikit-learn, Panda, Transformers (Hugging Face) Google Gemini API, pyttsx3 Technical Implementation\ Trained 3 ML models (Logistic Regression, Random Forest, XGBoost) on 74K+ tweets Multi-model debate combining VADER, TextBlob, and 3 transformer models (Twitter-RoBERTa, FinBERT, BERT) Feature engineering with TF-IDF vectorization, Word2Vec embeddings, and sentiment feature extraction Real-time inference pipeline for instant text analysis Cross-validation & evaluation with ROC curves, confusion matrices, and accuracy metrics Technical Implementation: Finetuned 3 Transformers models (BERT, finBERT,roBERTa) OCR Implementation (PyTesseract) Acronym explainer(Google Genai API) Text to Speech(streamlit,gtts,gTTS,io,BytesIo) Persona(PIL,pyteressact[need to download it incase you are running it] and google)

Built With

• panda
• python
• scikit-learn
• streamlit
• transformers-(hugging-face)-google-gemini-api