geronimo 0.3.0

ML deployment framework - from local development to production with one command

Geronimo: The Declarative ML Framework For AI

Build, train, and deploy ML models with production-ready infrastructure and Generative AI MCP support from the start.

Geronimo is like dbt for AI:

Why Geronimo?

🚀 Ship Models Faster

Stop writing boilerplate. One command creates a runnable project with FastAPI endpoints, monitoring, and CI/CD ready to go.

geronimo init --name iris-realtime
cd iris-realtime && uv sync
uvicorn iris-realtime.app:app --reload  # API running in seconds

🧩 Simpler Development

Define your model's what, not the how. The SDK has 5 components—each maps to one file:

Component	File	Purpose
DataSource	data_sources.py	Where your data comes from
FeatureSet	features.py	How to transform raw data
Model	model.py	Training and prediction logic
Endpoint	endpoint.py	Request/response handling (realtime)
Pipeline	pipeline.py	Batch job orchestration (batch)

data_sources.py — Declare your data

training_data = DataSource(name="training", source="snowflake", query=Query.from_file("train.sql"))

features.py — Define transformations

class IrisFeatures(FeatureSet):
    sepal_length = Feature(dtype="numeric", transformer=StandardScaler())
    sepal_width = Feature(dtype="numeric", transformer=StandardScaler())
    petal_length = Feature(dtype="numeric", transformer=StandardScaler())
    petal_width = Feature(dtype="numeric", transformer=StandardScaler())

model.py — Train and predict

class IrisModel(Model):
    name = "iris-realtime"
    features = IrisFeatures()
    
    def train(self, X, y, params): ...
    def predict(self, X): ...

endpoint.py — Handle requests (realtime)

class PredictEndpoint(Endpoint):
    def preprocess(self, request): ...
    def postprocess(self, prediction): ...

pipeline.py — Run batch jobs

class ScoringPipeline(BatchPipeline):
    schedule = Schedule.daily(hour=6)
    def run(self): ...

🤖 GenAI Agent-Ready

Every project is automatically exposed as an MCP tool. AI agents like Claude can call your models directly—no extra work required.

{
  "mcpServers": {
    "iris-realtime": {
      "command": "uv",
      "args": ["run", "python", "-m", "iris-realtime.agent.server"]
    }
  }
}

"Analyze this transaction for fraud risk"
→ Claude calls your model → Returns risk score

---

Getting Started

pip install geronimo
geronimo init --name my-model --template realtime

Choose your template:

Template	Use Case	Output
realtime	REST APIs, low-latency	FastAPI + monitoring
batch	Scheduled jobs, bulk scoring	Metaflow + drift detection
both	APIs + scheduled pipelines	Everything

What You Get

A complete, runnable project structure:

my-model/
├── src/my_model/
│   ├── sdk/                    # Define your model here
│   │   ├── model.py            # train() + predict()
│   │   ├── features.py         # Feature transformations
│   │   ├── endpoint.py         # Request/response handling
│   │   └── monitoring_config.py
│   ├── app.py                  # FastAPI (auto-generated)
│   └── train.py                # Training script
├── geronimo.yaml               # Deployment config
└── models/                     # Saved artifacts

Focus on the sdk/ folder. Everything else is generated for you.

Deploy to Production

geronimo generate all

Generates:

• Terraform — ECS Fargate infrastructure
• Dockerfile — Optimized for ML serving
• CI/CD — Azure DevOps / GitHub Actions pipelines

Integrations

Integration	Purpose
MLflow	Experiment tracking, artifact store
Snowflake/Postgres	Data sources for training
CloudWatch	Production metrics
Slack	Alerts for drift/errors
MCP	AI agent tool exposure

Documentation

• Getting Started: Realtime
• Getting Started: Batch
• Monitoring & Drift Detection
• MCP Integration
• SDK Reference

Installation

pip install geronimo                  # Core
pip install geronimo[mlflow]          # + MLflow
pip install geronimo[databases]       # + Snowflake, Postgres
pip install geronimo[all]             # Everything

---

Apache 2.0 License • GitHub