DOIN Forecast Provider — Decentralized ML Prediction Marketplace

Marketplace

Three roles, one platform

A complete marketplace connecting ML model providers with prediction consumers.

🛡️

Admin

Manage the platform, users, and billing.

User activation & role management
System configuration & monitoring
Billing oversight & API key management
Model approval & quality control

🧠

Provider

Register trained models (Keras, PyTorch, etc.)
Set pricing per prediction request
Monitor usage analytics & earnings
Update models with improved versions

📡

Client

Access verified ML predictions via API.

Browse available prediction models
Request predictions via REST API
Pay per request with transparent billing
Get confidence intervals & uncertainty estimates

Features

Production-grade prediction serving

Built for reliability, security, and scale.

🔌

Plugin Architecture

Every component is a plugin: core, endpoints, feeder, pipeline, and predictor. Swap any part without touching the rest.

🤖

Multi-Model Support

Run multiple ML models simultaneously. Keras, TensorFlow, PyTorch — any framework through the predictor plugin interface.

💰

Per-Request Billing

Transparent pay-per-prediction billing. Providers set prices, clients pay per request. Full accounting trail.

🧪

Noisy Ideal Testing

Validate model quality with synthetic noisy-ideal data. Ensures predictions meet accuracy thresholds before serving.

🌐

REST API

Complete FastAPI-based REST API with OpenAPI docs. Authentication, prediction requests, user management, model registry.

📊

Monte Carlo Uncertainty

Confidence intervals via MC dropout sampling. Know not just the prediction, but how confident the model is.

🔐

JWT Security

API key authentication, JWT tokens, role-based access control, user activation workflow. Production security out of the box.

⚙️

Configurable Logging

Fine-grained log levels via config, CLI, or PREDICTION_PROVIDER_QUIET=1 environment variable for silent operation.

API

Simple REST interface

Request predictions with a single API call.

# Authenticate
curl -X POST http://localhost:8000/auth/token \
  -d '{"username": "client1", "password": "secret"}'

# Request a prediction
curl -X POST http://localhost:8000/predictions/request \
  -H "Authorization: Bearer <token>" \
  -H "Content-Type: application/json" \
  -d '{
    "instrument": "EUR_USD",
    "prediction_horizon": 6,
    "model_id": "transformer_1h"
  }'

# Response
{
  "prediction": [1.0842, 1.0845, 1.0838, 1.0851, 1.0847, 1.0853],
  "confidence": 0.87,
  "uncertainty": [0.0012, 0.0015, 0.0018, 0.0021, 0.0024, 0.0027],
  "model": "transformer_1h_v3",
  "cost": 0.001
}

Getting Started

Running in 2 minutes

Install, configure, serve predictions.

# Clone and install
git clone https://github.com/harveybc/prediction_provider.git
cd prediction_provider
pip install -e .

# Run the server
python -m app.main --host 0.0.0.0 --port 8000

# Run in quiet mode
PREDICTION_PROVIDER_QUIET=1 python -m app.main

# Or with a config file
python -m app.main --load_config config.json --log_level WARNING

# API docs available at
open http://localhost:8000/docs

ML Predictions
as a Service.
Pay per request.

Three roles, one platform

Admin

Provider

Client

Production-grade prediction serving

Plugin Architecture

Multi-Model Support

Per-Request Billing

Noisy Ideal Testing

REST API

Monte Carlo Uncertainty

JWT Security

Configurable Logging

Simple REST interface

Running in 2 minutes

Serve predictions, earn rewards

ML Predictionsas a Service.Pay per request.

Three roles, one platform

Admin

Provider

Client

Production-grade prediction serving

Plugin Architecture

Multi-Model Support

Per-Request Billing

Noisy Ideal Testing

REST API

Monte Carlo Uncertainty

JWT Security

Configurable Logging

Simple REST interface

Running in 2 minutes

Serve predictions, earn rewards

ML Predictions
as a Service.
Pay per request.