Catagorys-API / README.md
ClergeF's picture
Update README.md
e0b9ba6 verified
metadata
title: Catagorys API
emoji: 
colorFrom: blue
colorTo: gray
sdk: docker
pinned: false
license: apache-2.0

MVT Community Value Model API

This API powers the MyVillage Token (MVT) scoring system.
It evaluates real-world community actions and predicts:


1. Estimated Dollar Value

How financially impactful the action is.

2. Impact Level

Overall significance of the action (0–1 scaled).

3. Category Strength Scores (0–1)

Each action is analyzed across 10 categories:

  • Family
  • Community
  • Education
  • Health
  • Environment
  • Business
  • Finance
  • History / Heritage
  • Spirituality
  • Innovation
  • Impact (standalone model)

These models were trained using a custom dataset of 2,000+ human-created community actions, labeled by the MVT team.


How the API Works

Each model is stored separately on the HuggingFace Hub.
This Space loads:

  • A universal text embedder
  • 12 linear regression models trained on SentenceTransformer embeddings
  • Outputs predictions for a single input sentence

API Usage

POST /predict

Input:

{
  "text": "I helped my neighbor fix her fence after the storm."
}

Output:

{
  "input": "I helped my neighbor fix her fence after the storm.",
  "predictions": {
    "estimated_value": 42.18,
    "impact_level": 0.67,
    "impact": 0.61,
    "family": 0.20,
    "community": 0.88,
    "education": 0.05,
    "health": 0.11,
    "environment": 0.22,
    "business": 0.03,
    "finance": 0.02,
    "history": 0.01,
    "spirituality": 0.04,
    "innovation": 0.08
  }
}

Running Locally

uvicorn app:app --reload --host 0.0.0.0 --port 7860

Then open:

http://localhost:7860/docs

Models Used

Each model is individually hosted under the ClergeF/ namespace:

  • value-impact-model
  • impact-model
  • family-model
  • community-model
  • education-model
  • health-model
  • environment-model
  • business-model
  • finance-model
  • history-model
  • spirituality-model
  • innovation-model

Embedder sourced from:

  • MVT-models/universal_embedder