Back office ops · Production
AppFolio builds Realm-X AI property management assistant using LangChain, LangGraph, and LangSmith
The problem
AppFolio needed a better natural language interface to help property managers engage with the platform and simplify operational processes, and as Realm-X evolved it required a way to handle greater complexity in multi-step requests.
Workflow diagram · grounded in source
1
User query via conversational interface
trigger
“Realm-X provides a conversational interface that helps users understand the state of their business, get help, and execute actions in bulk – whether it's querying information, sending messages, or scheduling actions related to residents,…”
2
Parallel branch execution
ai_action
“One major benefit of LangGraph has been its ability to run independent code branches in parallel. While determining relevant actions, it simultaneously calculates fallbacks and runs a question-answering bot over help pages. This allows R…”
3
Dynamic few-shot prompting
ai_action
“dynamic few-shot prompting — which involves dynamically pulling relevant examples to deliver more personalized and accurate responses to Realm-X users”
4
Realm-X produces actionable output
output
“helped Realm-X Assistant aggregate responses from multiple system parts to ensure clear, actionable user outputs”
5
Automated feedback and LLM evaluation
feedback_loop
“The team also added in automatic triggers to collect feedback when users submit an action drafted by Realm-X. In addition, automatic feedback is generated based on LLM or heuristic evaluators to continuously monitor system health.”
Reported outcome
Early Realm-X users save over 10 hours a week, and text-to-data accuracy improved from roughly 40% to 80% after dynamic few-shot prompting was introduced.
AppFolio has also maintained high performance as they expanded the number of actions and data models available.
Reported metrics
Property manager time saved per weekover 10 hours a week
Text-to-data accuracy~40% to ~80%
Reported stack
LangChainLangGraphLangSmith
Frequently asked questions
What did this team achieve with this AI workflow?
Early Realm-X users save over 10 hours a week, and text-to-data accuracy improved from roughly 40% to 80% after dynamic few-shot prompting was introduced.
What tools did this team use?
LangChain, LangGraph, LangSmith.
What results were reported?
Property manager time saved per week: over 10 hours a week; Text-to-data accuracy: ~40% to ~80% (source-reported, not independently verified).
How is this back office ops AI workflow structured?
User query via conversational interface → Parallel branch execution → Dynamic few-shot prompting → Realm-X produces actionable output → Automated feedback and LLM evaluation.