A meta-learning platform that treats prior experiments as training data. Teach machines how to learn, not just what to learn.
Orca is a monorepo for meta-learning. The core premise: past experiments carry reusable signal, and a system that remembers them should outperform one that starts from scratch every time. Orca embeds ML tasks into a vector space, tracks what worked before, and uses that history to recommend models, warm-start training, and steer hyperparameter search.
The platform has three services and a shared infrastructure layer:
| Component | Codename | Role |
|---|---|---|
| OrcaMind | The Brain | Meta-learning engine: task embedding, model selection, MAML/Reptile/Meta-SGD, warm-start transfer |
| OrcaLab | The Lab | Experiment management: adaptive hyperparameter search, Prefect orchestration, live dashboards |
| OrcaNet | The Connector | Cross-domain knowledge transfer: domain-invariant embeddings, LLM-powered reasoning, transfer scoring |
| orca-shared | The Foundation | Shared schemas, SQLAlchemy ORM, storage backends, MLflow wrappers, HTTP client library |
For prerequisites and local dev setup, see Getting Started.
git clone https://github.com/AruneemB/orca.git
cd orca
# Start backing services
docker compose -f docker-compose.dev.yml up -d postgres redis minio mlflow
docker compose -f docker-compose.dev.yml run --rm orcamind python scripts/init_db.py
docker compose -f docker-compose.dev.yml up -d orcamindOr with Make:
make install
make docker-up| Guide | Description |
|---|---|
| Getting Started | Prerequisites, Docker Compose setup, local dev mode |
| Components | orca-shared and OrcaMind internals, API, CLI, dashboard |
| Architecture | System diagram, repo layout, tech stack |
| Database | Alembic migrations, OpenML meta-dataset seeding |
| Development | Testing, linting, type checking, pre-commit, Makefile |
| Deployment | Environment variables, service topology, production notes |
| API Reference | REST endpoint specs for all three services |
| Roadmap | Planned features, reference papers |
| Packages | Package-level READMEs for orca-shared, OrcaMind, OrcaLab, OrcaNet |
| Scripts | Operational scripts: database migrations, Prefect work-pool setup, OpenML meta-dataset seeding |