Palantir Reference Architecture

Overview

Production-ready architecture for Foundry-integrated applications. Covers the standard data pipeline pattern (ingest > clean > model > serve), Ontology design, external API integration, and multi-repo project layout.

Prerequisites

• Foundry enrollment with project access
• Understanding of Ontology concepts (object types, link types, actions)
• Familiarity with palantir-core-workflow-a (transforms) and palantir-core-workflow-b (Ontology)

Instructions

Step 1: Data Pipeline Architecture

┌─────────────┐     ┌──────────────┐     ┌─────────────┐     ┌───────────┐
│  Raw Layer   │────>│  Clean Layer │────>│ Model Layer │────>│ Ontology  │
│ (ingested)   │     │  (validated) │     │ (enriched)  │     │ (objects) │
└─────────────┘     └──────────────┘     └─────────────┘     └───────────┘
  ↑ Connectors        @transform_df       @transform_df       Object types
  ↑ REST sync          null checks         joins, aggs         Link types
  ↑ File upload        type casting        ML features         Actions

Step 2: Project Layout (Foundry)

Foundry Project: "Customer Analytics"
├── Datasets/
│   ├── raw/                    # Ingested from sources
│   │   ├── raw_orders          # REST connector → CRM
│   │   ├── raw_customers       # JDBC connector → DB
│   │   └── raw_products        # File upload (CSV/Parquet)
│   ├── clean/                  # Validated, typed
│   │   ├── clean_orders        # Nulls removed, dates parsed
│   │   ├── clean_customers     # Deduped, normalized
│   │   └── clean_products      # Schema enforced
│   └── model/                  # Enriched, analytics-ready
│       ├── order_enriched      # Joined with customer + product
│       ├── customer_360        # Aggregated customer view
│       └── daily_summary       # Time-series aggregation
├── Code Repositories/
│   ├── pipeline-ingestion/     # Connectors and raw → clean
│   ├── pipeline-analytics/     # Clean → model transforms
│   └── ontology-actions/       # Action implementations
└── Ontology/
    ├── Object Types: Customer, Order, Product
    ├── Link Types: Customer→Orders, Order→Products
    └── Actions: createOrder, updateCustomerSegment

Step 3: External API Integration Pattern

# External app consuming Foundry Ontology via Platform SDK
my-external-app/
├── src/
│   ├── foundry/
│   │   ├── client.py           # Singleton FoundryClient
│   │   ├── objects.py          # Object query helpers
│   │   ├── actions.py          # Action wrappers
│   │   └── cache.py            # TTL cache layer
│   ├── api/
│   │   ├── routes.py           # REST endpoints
│   │   └── webhooks.py         # Foundry event handlers
│   └── main.py
├── tests/
│   ├── conftest.py             # Mocked FoundryClient
│   ├── test_objects.py
│   └── test_actions.py
├── .env                        # FOUNDRY_HOSTNAME, credentials
└── requirements.txt

Step 4: Ontology Design Patterns

Step 5: Security Layers

┌──────────────────────────────────────────┐
│ Layer 1: Network (VPN/private link)       │
├──────────────────────────────────────────┤
│ Layer 2: OAuth2 (service user per app)    │
├──────────────────────────────────────────┤
│ Layer 3: Scopes (minimum per app)         │
├──────────────────────────────────────────┤
│ Layer 4: Project roles (Viewer/Editor)    │
├──────────────────────────────────────────┤
│ Layer 5: Marking (data classification)    │
└──────────────────────────────────────────┘

Output

• Standard 3-layer data pipeline (raw > clean > model)
• Ontology design with typed objects, links, and actions
• External app architecture with caching and webhooks
• Security model with 5 defense layers

Error Handling

Resources

• Foundry Documentation
• Ontology SDK Overview
• Transforms Guide

Next Steps

For data handling and compliance, see palantir-data-handling.