Back to Projects
01Completed 2025

Multimodal Sentiment Analysis

Built a production-grade AI system from scratch that quantifies emotions from short video content using multimodal analysis.

Multimodal Sentiment Analysis

Overview

An end-to-end sentiment analysis platform that processes video, audio, and text simultaneously. The system breaks videos into utterances, analyzes each with deep learning models, and provides granular emotional insights through a modern dashboard.

Key Features

  • 01Multimodal AI: Analyzes Video, Audio, and Text simultaneously for higher accuracy
  • 02Granular Analysis: Breaks down videos into specific utterances (sentences) and analyzes each one
  • 037 Emotion Classes: Detects Anger, Disgust, Fear, Joy, Neutral, Sadness, and Surprise
  • 04Sentiment Detection: Classifies content as Positive, Neutral, or Negative
  • 05Developer API: Provides a secure API with quota management for developers
  • 06Modern Dashboard: A clean Next.js interface to upload videos and view detailed results

Impact

Trained on 10k samples

Technologies

Frontend

Next.js (App Router), Tailwind CSS, NextAuth.js, TypeScript

Backend

PyTorch, OpenAI Whisper, AWS SageMaker, AWS S3

AI / ML

Multimodal Deep Learning

Database

PostgreSQL (Prisma)

Tools

FFmpeg, OpenCV

PreviousFootball Turf AnalyticsNext Manaura Fitness