WhereTF

WhereTF is an AI-powered file search tool that lets users find files using natural language and contextual memory instead of exact filenames. It understands file content and retrieves the most relevant results based on meaning, making file search faster and more intuitive.

---

Features

• Semantic searching across multiple file types
• Multimodal image & text vectorization
• Hybrid searching (semantic + keyword matching)
• Query Expansion (via NLTK/WordNet)
• Automated Indexing
• Relationship Mapping

Dependencies

• Docker
• Docker Compose
• PostgreSQL (with pgvector extension)

Configuration (App Tiers)

WhereTF supports multiple performance tiers so you can run it on everything from a standard laptop to a dedicated server. You can change your tier by modifying the APP_TIER environment variable in your docker-compose.yml.

Tier	RAM Usage	Text Embeddings	Vision Embeddings	EasyOCR
lite	~700 MB	Yes (all-MiniLM-L6-v2)	No	Yes
balanced	~1.2 GB	Yes (nomic-embed-vision-v1.5)	Yes (nomic-embed-vision-v1.5)	No
pro (Default)	~2.0 GB	Yes (jina-clip-v1)	Yes (jina-clip-v1)	Yes

• Lite: Best for older hardware. Uses a lightweight text model and relies entirely on OCR to read images.
• Balanced: Best middle-ground. Disables OCR to save PyTorch overhead, but enables state-of-the-art visual concept search using Nomic's highly efficient open-source vision model.
• Pro: Full feature set. Vectorizes text, visual concepts, AND extracts tiny text hidden inside diagrams and receipts using Jina CLIP and EasyOCR.

Hardware Acceleration (CPU vs GPU)

By default, WhereTF builds a highly optimized CPU-only Docker image to save disk space (~150MB instead of 2.5GB of Nvidia drivers).

If you have an Nvidia GPU and want to accelerate indexing and searching:

1. Open docker-compose.yml.
2. Under the build block for both the backend and worker services, set USE_GPU: 1.
3. Uncomment the deploy block in the compose file to pass GPU hardware capabilities to the container.
4. Run the build command below.

Running

First, clone the repository and navigate into the project directory:

git clone [https://github.com/GDGVIT/WhereTF-backend](https://github.com/GDGVIT/WhereTF-backend)
cd WhereTF-backend

Start the application and database using Docker Compose. The --build flag ensures your chosen GPU/CPU architecture is compiled correctly:

docker-compose up --build -d

(The backend will be available on port 8000, and the database will be mapped to port 5433)

To stop the services:

docker-compose down

Contributors

Maneet Gupta GitHub | LinkedIn

Aryan Rangarajan GitHub | LinkedIn

Prakhar Sethi GitHub | LinkedIn

Made with ❤ by GDSC-VIT