Multilingual Virtual Research Assistant (MVRA)

Generative AI
Large Multimodal Model
Image Analysis
Image Grounding
Llama4
Streamlit
Author

Prasad Bhoite

Tip

MVRA is a Streamlit-based GenAI assistant that empowers researchers with natural language understanding, multilingual translation, image analysis, and synthetic data generation β€” powered entirely by Meta’s LLaMA 4.

πŸš€ Features

  • πŸ“ Ask a Question: Natural language QA using LLaMA-4 Instruct models such as Llama 4 Scout and Llama 4 Maverick.
  • πŸ“° Summarize Text: Generate concise, research-oriented summaries.
  • πŸ–ΌοΈ Image Analysis (URL & Upload): Ask questions about images or compare multiple visuals.
  • 🌐 Multilingual Translator: Bi-directional translation between English, French, Spanish, Hindi, and Chinese.
  • 🧩 Image Grounding: Detect and visualize tools/items in images using bounding boxes.

🧱 Tech Stack

  • Frontend: Streamlit
  • LLMs: Meta LLaMA 4 (Scout/Instruct via llama-api)
  • Multilingual API: llama4 w/ compatibility base URL
  • Vision: Base64 encoding + VQA using multimodal LLaMA
  • Image Grounding: Regex + pydantic + matplotlib overlays
  • Deployment: Streamlit Cloud

πŸ”— GitHub Repository

πŸ‘‰ Explore Source Code: https://github.com/prasadbhoite/Multilingual_Virtual_Research_Asistant

πŸ§ͺ Screenshots

⚠️ If you’re running locally, make sure to configure your .env file and requirements.txt.

🌍 Live Demo

Try the app live:

```{=html} πŸ“š