Sentira AI is a real-time emotion detection system using Tri-Modal Neural Fusion. It processes acoustics, Whisper text, and RoBERTa semantics to detect "emotional masking." Unlike basic models, it accurately identifies high-energy accents often mislabeled as anger. Built with FastAPI and Docker, it offers near-zero latency and 99.33% accuracy. Sentira bridges the gap between how we sound and what we truly mean.

DigitalOcean Serverless Inference — 55+ AI models behind one OpenAI/Anthropic-compatible API

55+ AI models behind one OpenAI/Anthropic-compatible API

Hi everyone! 👋

I built Sentira AI to solve a problem I saw in most emotion AI: they are easily fooled. Most systems only listen to tone, missing the actual meaning of the words—or vice-versa.

I developed a Tri-Modal Neural Fusion system that processes acoustics (MFCC), transcription (Whisper), and deep semantics (RoBERTa) simultaneously. This allows the AI to detect "emotional masking" and handle high-energy accents that Western-trained models often mislabel as anger.

It’s live, containerized with Docker, and achieves 99.33% accuracy. I’d love for you to try and "outsmart" the AI with your best poker-voice! 🚀

Try the demo:

https://ds-portfolio-gules.vercel.app/emotion-demo