Palantir Hands-On Bootcamp: Property Maintenance Operations
Build an AI-powered property maintenance scheduling system from scratch using Palantir Foundry. In 47 lessons spanning a Foundation Track and an Advanced Track, you will upload data, build pipelines, model an Ontology, create Workshop dashboards, deploy AIP agents, build external apps with OSDK, and tie it all together with predictive maintenance — covering the full Foundry stack with bilingual content (English + 中文).
Premium content
Upgrade to premium to access all lessons, tutorials, and hands-on exercises.
Lesson Overview
The Palantir Story — From Seeing Stones to Enterprise AI
Meet Palantir Technologies — the company behind Foundry. Learn how a startup named after Tolkien's seeing stones grew from intelligence work into the world's leading data operating system for enterprises.
What is Palantir Foundry?
Understand what Palantir Foundry is, how it differs from traditional data tools, and meet the rental property management scenario we will build throughout this bootcamp.
Uploading & Exploring Data
Learn how to create a project folder, upload CSV datasets into Foundry, navigate the file browser, and explore dataset schemas and previews.
Pipeline Builder & Transformations
Use Foundry Pipeline Builder to visually join, filter, rename, add computed columns (expressions and CASE logic), and aggregate data — with a complete operations reference covering all Pipeline Builder node types.
What is the Ontology?
Understand the paradigm shift from rows in tables to objects that mirror the real world, and why the Ontology is the foundation of everything in Palantir Foundry.
Creating Object Types
Create three Ontology object types — Property, Work Order, and Technician — by mapping your CSV datasets to structured objects in Foundry.
Links, Actions & Object Explorer
Connect your object types with link types, build action types with validation rules, and explore your fully connected Ontology in Object Explorer.
Introduction to Workshop
Learn what Workshop is and how it enables building powerful applications on top of the Ontology without writing frontend code.
Building the Maintenance Dashboard
Build a complete interactive maintenance dashboard with property list, filtered work orders, detail view, and action buttons using Workshop.
Dashboard Charts & Analytics
Add interactive charts and analytics to the maintenance dashboard, including bar charts, line charts, pie charts, and a polished multi-tab layout.
AIP: AI Meets Your Data
Introduce Palantir AIP and demonstrate how operational AI queries your Ontology through natural language using AIP Agent Studio.
Building an AIP Agent
Build a Property Manager Assistant agent in Agent Studio that can read Ontology data and trigger Actions through natural language commands.
AIP Logic, Automate & Full Integration
Create an AIP Logic function for auto-classification, set up Automate triggers, embed the agent in Workshop, and walk through the complete end-to-end operational flow.
Foundry Developer Experience (FDE)
Learn how to use Code Repositories to replicate Pipeline Builder transforms in TypeScript, build a custom work order routing function, write unit tests, and publish versioned code alongside visual pipelines.
Data Lineage & Governance
Explore Foundry lineage graphs to trace data from raw CSV to AIP agent, perform impact analysis, configure health checks, use branching for safe changes, and set up granular permissions for compliance.
OSDK — Build External Apps with the Ontology
Build a tenant-facing React portal using the Ontology SDK (OSDK). Learn to register a third-party app, scaffold a React + OSDK project, query Ontology objects with type-safe TypeScript, apply actions, and deploy an external web app that consumes your Foundry data.
Capstone Project — Full-Stack Foundry
Apply everything from Lessons 1-16 in a comprehensive capstone: upload 6 realistic datasets, build pipelines, model 5 object types, create a scheduling dashboard, deploy an AIP agent, run predictive maintenance scoring in FDE, build an OSDK tenant portal, and govern the entire system.
Work Order Scheduling in Workshop
Build a 3-panel Workshop module for scheduling work orders — a detail panel showing work order metadata, a ranked appointment slot list with scoring metrics, and a Mapbox map pinning asset locations. Learn ontology design for scheduling, cross-panel variable wiring, conditional formatting, and map layers.
VROOM + OSRM Routing Engine
Deploy VROOM and OSRM locally via Docker, map Foundry ontology data to VROOM concepts (vehicles, jobs, skills, time windows, priorities), call the VROOM API to produce optimized technician routes, and write results back to the ontology. Compare automated routing quality against manual dispatch.
Automated Scheduling Pipeline + Capacity Engine
Build a production batch optimization pipeline (Code Workflow → Lambda → VROOM → writeback), implement constraint validation, create a capacity engine that computes bookable slots from raw capacity minus booked minus emergency reserves, handle mid-day emergency re-optimization, and deploy a quality metrics dashboard with shadow-mode testing.
OSDK Mobile Apps — Technician & Resident Portals
Build two OSDK-powered mobile applications: a Technician Portal showing today's optimized route with turn-by-turn navigation, real-time status updates, and push notifications; and a Resident Portal for viewing available slots, booking appointments, tracking technician ETA, and submitting maintenance requests. Connect both apps to the Foundry ontology for real-time data flow.
Capstone 2 — Autonomous Dispatch System
Tie together VROOM routing, the capacity engine, mobile apps, and the scheduling dashboard into a fully autonomous dispatch system. Build a comprehensive testing strategy with scenario libraries, implement graduated rollout (shadow → assisted → supervised → autonomous), and add AI enhancements including demand forecasting, dynamic emergency reserves, and auto-dispatch rules.
Building a Custom Widget for Workshop
Build a custom React + TypeScript widget for Workshop — a Maintenance Progress Card driven by primitive parameters — then publish it and wire it to live Ontology data. No special permissions required.
Custom Widgets 2 — Interactive Tables & Multi-Select
Go beyond display widgets: build a custom widget that reads a live Ontology objectSet, renders a checkbox multi-column table, and writes the selection back to Workshop to drive the rest of the dashboard. Covers the objectSet parameter, the Enable-Ontology-APIs permission gate, OSDK rendering, and cross-widget interactivity.
Measuring AI with AIP Evals
Score your LLM functions instead of guessing. Build an evaluation suite, grade a Q&A bot with an LLM-as-a-judge, read the pass rate, compare models, and debug the classic JSON-parse failure.
Exploring Data with AIP Analyst
Ask your data a question in plain language and let AIP Analyst search the Ontology, build object sets, and chart the answer — no pipeline, no SQL. Learn the Settings that drive accuracy and when to explore vs. build.
AI FDE — Building Foundry by Conversation
AI FDE is an agent that operates Foundry through natural language — building pipelines, ontology, functions, and apps. Learn the 9 Modes, the tiered safety model, and how to make it reliable with AIP Skills.
AI FDE Skills — a Reusable Troubleshooting Library
Ten reusable AIP Skills that make AI FDE reliable — a Foundry data-ops playbook. Covers SNAPSHOT uploads, build order, Merge vs Deploy vs Build, composite primary keys, build-failure triage, and the verify-data-freshness meta-skill (four green stages are not the same as latest). Each skill's Name / Description / Markdown.
Auto-Ingest & Refresh — Hands-off Data Updates
Build the chain where uploading a file automatically reruns the pipeline and refreshes Workshop — no manual build. Covers SNAPSHOT (replace, not append) end-to-end, the Workshop upload entry, the event-triggered Schedule (trigger on input, not cron) with build-required-inputs, and object-type auto-reindexing.
Reading Foundry Data from the Command Line
Pull Ontology objects and datasets headless — no UI. Mint a client-credentials token, then read via the v2 REST API (curl/fetch: list, select, orderBy, search, paginate), the OSDK, or the SQL Query API — the pattern a real production service uses.
Writing to Foundry via Actions
Create and update Ontology objects headless with the v2 Apply Action API. Learn why writes go through Actions, the critical gotcha that a 200 is NOT success (check validation.result), how to dry-run with VALIDATE_ONLY, and how returnEdits gives you the new primaryKey.
OSDK — the Typed SDK
Use the Ontology SDK (OSDK) to query and write Foundry with fully typed objects and actions. Set up createConfidentialOauthClient + createClient, run typed .where().fetchPage() queries with pagination, and apply actions with .applyAction() that throws on rejection. The typed door, alongside raw REST (L30) and Actions (L31).
Auto-Alert: Analyst → Logic → Automate
Build a live, no-code automation end to end — AIP Analyst proves the data, an AIP Logic function returns ontology edits, and an Automate rule watches a region's revenue and writes a 落后 status back on its own. The no-code sibling of the Actions API write path (L31), plus the traps that waste the most time.
Global Branching & Proposals — One Change, Three Apps
Put a pipeline, an object type and a dashboard change on one global branch, review them as one proposal, rebase when Main moves underneath you, and merge them together — built and merged for real in the Training Sandbox, with the traps that cost the most time.
Code Repositories CI/CD — Protected Branches, Red Builds, Tagged Releases
Protect master, let a unit test fail a pull request, fix it, get it approved, squash-merge it and tag a release — then see why a stable version can only come from a protected branch. Built live in the Training Sandbox on Lesson 14's routing repo.
Automate Part 3 — Chaining, Retries & Failure Monitoring
Chain a second automation off Lesson 33's alert with an Automation dependency, read the two layers of retries, fallback actions and auto-mute, catch an automation re-triggering itself, and put monitoring rules on the chain in Data Health — built and run live in the Training Sandbox.
Workshop Advanced — Embedded Modules, Routing, State & Events
Turn a dashboard into an app: a child module with a typed interface embedded in the parent, a section that hides itself when empty, a filter carried in the URL, saved viewer states, inline edits through an Action, and a row click that opens another module — built live on a duplicate in the Training Sandbox.
Document Intelligence — From Scanned PDFs to Searchable Chunks
Pull text, tables and checkboxes out of PDFs with AIP Document Intelligence: compare Raw text, OCR, Layout-aware OCR and two VLMs on a noisy scan, let a VLM grade each result, preview chunking for RAG, and deploy the winner as a transforms repository — including the two real build errors and their fixes.
Quiver — Point-and-Click Analysis, Dashboards & a Workshop Embed
Analyse Rental Work Orders in Quiver: a monthly trend line by category, a filter that exposes dirty priority values, a search-around to technicians, a bar chart with drill-down, then a published dashboard embedded in a Workshop module — built live in the Training Sandbox.
AIP Logic Part 2 — Loops, Reduce, Conditionals & Tools
Score a whole batch of work orders in one AIP Logic function: a Loop runs an LLM per item, Reduce folds the scores into one number, and a Conditional turns it into ESCALATE or ROUTINE — plus structured-output modes, tools, custom tools, timeouts, inline code and four real type errors, built live in the Training Sandbox.
AIP Evals Part 2 — Experiments, Branches & Run History
Grade a published AIP Logic function with an evaluation suite, then go further: pin the version under test, repeat tests to expose unstable answers, compare runs, sweep a threshold as an experiment on a branch, and learn where results live — including a 0% run caused by the grader, not the model, all captured live in the Training Sandbox.
Managing an Ontology Over Time — Value Types, Shared Properties & Cleanup
Keep a large shared ontology healthy: define valid values once with a value type, reuse meaning with a shared property, attach it to an object type on a branch, and use Health issues and Cleanup to find what is broken or unused — plus permissions, interfaces and migration, all captured live in the Training Sandbox.
Schedules Part 2 — Failure Handling: Expectations, Retries & Health Checks
Break a scheduled build on purpose and follow the failure everywhere: a data expectation fails the build, the dataset keeps its last good data, the schedule and its health check turn red — then fix it, meet the up-to-date trap and Force build, and tidy up. Retries, build scope and schedule versions, all captured live in the Training Sandbox.
Compute Modules — Run VROOM Inside Foundry as a Function
Move the Lesson 19 route solver inside Foundry: a Functions-mode compute module, a Python Compute Module repository with pyvroom, an image built by Tag version (no Docker login), a container, an imported function and real queries — plus a failure that reported success, and the fix. Captured live in the Training Sandbox.
Release to Production — DevOps, Marketplace & the Admin-Owned Blockers
A theory lesson from Palantir's documentation: environments as spaces, Foundry DevOps products and stores, Marketplace installations, installation modes, release channels, recall and rollback, the checks to run before release — and the admin-owned settings (kiosk, CSP, CORS, egress, allowlists, markings) that can block a perfect release.
MCP Hub — Palantir MCP vs Ontology MCP, Tokens & Governance
A docs-only lesson: what MCP Hub is, Palantir MCP for builders (ontology structure, via proposals) versus Ontology MCP for consumers (data, via actions), whose permissions every tool call runs with, where the data goes, Palantir as an MCP client through pro-code agents — and why a token never belongs on screen.
AIP Evolve — Agents That Improve Your AI Systems (and How to Review Them)
A docs-only lesson: AIP Evolve coordinates fleets of AI FDE agents to improve AI systems. Set a target, goal, validation strategy and constraints; read the agent graph and the proposal from the docs' worked example; review validation, confidence and limitations before merging through Global Branching — plus cost, permissions and the gaps in the docs.