Apppublic

cj-dev-code/semantic_search

sourceHugging Faceupdated 1y agoView on Hugging Face

0likes

App README

🧠 Mental Health Semantic Searcher

A project to make mental health advice from dense books more accessible through semantic search and quote-level retrieval.

✨ Motivation

When you're dysregulated, skimming a whole book or even a chapter can feel overwhelming. This tool helps you search for therapeutically helpful language based on what you're feeling or trying to express — so you can find quotes, not just concepts.

⚙️ Tech Stack

—Python – backend logic + semantic chunking
—Azure AI Search – vector search. rejected: Pinecone, FAISS, Vilvus
—Vector Embeddings - voyage 3.5. chosen for: intent, emotional, concept rich embedding. rejected: openai (too literal), cohere (unemotional), MPnet MiniLM (poor purpose handling), Google USE (too generic for quote level lookup)
—Gradio – frontend interface. Options rejected: streamlit, dash, next.js, react.
—GitHub Actions – CI pipeline
—pytest – test framework

📂 Project Structure

frontend/    # User interface (Streamlit or React)
backend/     # API + semantic search logic
data/        # Sample books and chunked quote data
notebooks/   # R&D, experiments, visualizations
scripts/     # Setup scripts, evaluators

🚧 Status

—[x] Repo initialized with CI
—[x] Project scaffolded
—[ ] Semantic chunker implemented
—[ ] Retrieval working end-to-end
—[ ] Frontend MVP

🧪 Running Tests

bash

pip install -r requirements.txt
pytest

---

## 🔐 Environment Variables

Before running the app, create a `.env` file in the project root and define the following:

| Variable           | Description                          |
|--------------------|--------------------------------------|
| `VOYAGE_API_KEY`   | API key from [Voyage AI](https://voyageai.com) |
| `AAIS_ENDPOINT`    | Azure AI Search endpoint URL         |
| `AAIS_KEY`         | Azure AI Search admin key            |
| `OPENAI API KEY`   | Openai API key                       |
| `RUN_DOCKER_TEST`  | make this true to run docker tests   |

You can use the provided `.env.example` file as a template: