Quickstart Guide: 0 to Production Model

Summary: In this quickstart guide, you will learn the core Doushi workflow: creating a project, uploading a tabular dataset, writing a plain-English prediction prompt, testing with interactive sliders, and obtaining an HTTPS API key.

Prerequisites

• A Doushi account (sign up free at app.doushi.ai)
• A tabular dataset in CSV, Excel (.xlsx), or Parquet (.parquet) format (or use one of our built-in sample datasets)

---

Step 1: Create a Project & Upload Data

1. Log into your dashboard at app.doushi.ai.
2. Click New Project.
3. Drag and drop your dataset into the upload zone.

Doushi will immediately run an initial profile scan:

• Column data types (numerical, categorical, datetime, text)
• Row and column counts
• Missing value percentage per column

---

Step 2: Define Your Target with a Natural Language Prompt

In the prompt input box, describe what you want the model to predict:

"Predict whether a customer will churn (column: churned) using their tenure, 
monthly spend, support ticket frequency, and contract type. Optimize for high precision."

You can also include business constraints:

• "Ignore customer_id and signup_timestamp"
• "Penalize false positives more heavily than false negatives"
• "Provide top 3 feature drivers for every prediction"

Click Generate Model.

---

Step 3: Automated Sandbox Training

Doushi spins up an isolated Docker container in our Frankfurt am Main (Germany) cluster. Within 60 seconds:

1. Missing values are imputed using robust median and mode strategies.
2. Categorical columns are frequency-encoded or target-encoded.
3. Leakage detection strips out features suspiciously correlated with the label.
4. Model Tournament evaluates LightGBM, XGBoost, CatBoost, and Random Forests with 5-fold cross-validation.
5. The champion model is locked and deployed.

You will see the final evaluation metrics:

• ROC-AUC Score (e.g., 0.932)
• F1-Score (e.g., 0.884)
• Accuracy (e.g., 91.6%)
• Average Latency (e.g., 12.8ms)

---

Step 4: Validate in the Live Interactive Simulator

Before writing code, test your model directly in your browser:

• Adjust numerical input sliders (e.g. monthly_spend, tenure_months).
• Select values from dropdown menus (e.g. contract_type).
• Observe the real-time prediction score (e.g. 82% Churn Probability) update in under 75 milliseconds.
• View Feature Importance Badges explaining what drove the decision.

---

Step 5: Integrate via REST API

Navigate to the API & Integration tab. Your production HTTPS endpoint is already active:

curl -X POST https://api.doushi.ai/api/projects/<YOUR_PROJECT_ID>/predict \
  -H "X-API-Key: <YOUR_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "monthly_spend": 79.50,
    "tenure_months": 8,
    "support_tickets": 3,
    "contract_type": "month-to-month"
  }'

Response:

{
  "status": "success",
  "prediction": 1,
  "probability": 0.764,
  "latency_ms": 13.2
}

---

Next Steps

• Review the REST API Reference for batch and streaming prediction endpoints.
• Learn how to use the Python CLI Guide to automate training in CI/CD.
• Read about our Docker Sandbox Security and GDPR compliance.