Why your data foundation determines everything your agents can do
Every AI agent, every automation, every autonomous workflow starts in the same place: clean, validated, connected building data. Without it, agents have nothing to reason over.
This session of the Agentic Proptech Webinar Series goes deep on onboarding — what it actually means to get your buildings, systems, and sensor data modeled in RealEstateCore, and how a process that once took months now takes hours.
In 50 minutes, Per Karlberg (CEO) and Felicia Helldén Rhodin (Customer Success Manager) walk through what onboarding involves end-to-end, demonstrate a live AI-powered onboarding of a real building with a real tag list, and show the Systema Machinae model that makes it possible. If you are planning any agentic work in your portfolio, this is the session to watch first.
What this session covers
- What onboarding actually means in an agentic proptech context — not inviting users to a system, but getting your BMS, FM, IoT, BIM, and energy data modeled, connected, and validated in RealEstateCore
- Why BMS systems are the hardest part — thousands of locally named, poorly standardized tags across buildings, countries, and vendor systems, and how ProptechOS solves this at scale
- Live demo: onboarding a real building in real time — a full tag list, an empty building in ProptechOS, and an onboarding agent running a 16-step process from four lines of input
- Systema Machinae explained — the custom-trained embedding model built from millions of hand-modeled tags that automatically maps sensor tags to RealEstateCore with a confidence score
- How the onboarding agent works — from reading the tag list and running Systema Machinae, to reviewing low-confidence tags, creating digital twins, and validating the result
- What actually slows onboarding down today — it is no longer the modeling work; it is system access, networking permissions, and VPN setup — and how to get ahead of it
- Real results — 15 buildings with two BMS systems each onboarded in three days; a homogeneous portfolio of 17 buildings onboarded in three hours
What you will see in the live demo
Systema Machinae: automated tag mapping A real tag list of 3,500 previously unseen tags is uploaded to System Machina. The model places each tag into a 1,536-dimension embedding space built from millions of hand-modeled examples, returning a confidence score for each one. High-confidence tags are automapped instantly. Lower-confidence tags are flagged for review. In this demo, 2,800 of 3,400 tags are automapped without any human intervention.
End-to-end onboarding agent A second demo shows what onboarding looks like when an agent runs the entire process. Input: a tag list, a building name, and a task in Jira. The agent reads the tag list, identifies the system type and tagging standard, runs Systema Machinae, reviews low-confidence results using its own reasoning, creates all digital twins in ProptechOS, places sensors in the correct rooms, and validates the output — logging every step back to the project management system.
Zero standing privileges in practice Before the onboarding agent begins, it has no access to ProptechOS. It explains to the ProptechOS permission layer what it needs to do and points to the Jira task as verification. It receives a temporary, scoped access — write permission for digital twins in that specific building only — and that access is revoked automatically when the work is complete.
How Systema Machinae works
ProptechOS has hand-modeled millions of BMS tags since 2018 — across five continents, dozens of vendor systems, and multiple tagging standards and languages. The first building took almost a month. Within a year, internal tooling brought that to a week. Within a few years, to a few days.
Systema Machinae is what was built from that history.
It is a custom text embedding model trained from scratch on a curated sample of those millions of modeled tags. The model creates a map — not in two dimensions like longitude and latitude, but in 1,536 dimensions — where every known tag occupies a specific location in the embedding space. Points close together in that space share meaning. Points far apart do not.
When a new tag list arrives, each tag is placed into that space. The model finds the nearest known neighbors and returns a confidence score. Tags with close neighbors are automapped. Tags in sparse areas of the space are flagged for human or LLM review.
No hand-crafted rules. No manual logic. The model has developed an intuition for what a tag means — because it learned from millions of examples of tags that were modeled correctly by domain experts.
What onboarding looks like in practice today
The balance of work has shifted completely. What was previously 80% modeling work and 20% coordination with customers and partners is now the inverse — 95% of the effort is getting system access, understanding networking and VPN constraints, and identifying the right contact at the building vendor or system maintainer. The actual onboarding work has shrunk to a small fraction of the project.
What remains for the customer success and onboarding team is validation — reviewing what the agent produced, confirming sensors are placed correctly, and catching anything the model could not resolve on its own.
ProptechOS assigns a dedicated onboarding engineer to every customer account. They track the data, know the buildings, and handle validation alongside the customer success manager. The onboarding engineer is the constant throughout the process.
What you need to get started
API documentation Only required if your system is not already among the 90+ connectors ProptechOS has built. If a connector exists, documentation is not needed.
System access Login credentials so ProptechOS can connect to the data in your buildings. This is now the most critical and most commonly delayed input. Getting access early is the single biggest success factor.
A tag list The list of sensor points, devices, and assets to be onboarded. This can be provided as a file or fetched directly from the system UI. The agent will work through the list to determine what to onboard and what to leave out.
A contact person Either someone at your organization or a direct contact at the system vendor. Direct developer-to-developer contact when questions arise makes a significant difference in onboarding speed and quality.
What can be onboarded into ProptechOS
- BMS systems — the most common and historically the most complex, now handled by System Machina and the onboarding agent
- FM systems — work orders, alerts, service objects
- IoT systems — sensors, devices, actuators
- BIM models — 3D building structure, room names, spatial placement of assets
- Energy and metering systems
- Business and ERP data
Once modeled in RealEstateCore, data from any of these systems — regardless of vendor, country, or tagging standard — becomes directly comparable across your entire portfolio.
See what your buildings look like in ProptechOS
Book a demo and we can walk through what onboarding your portfolio would involve and which systems you could connect first.
Frequently asked questions
Q: What is a digital twin in the ProptechOS context? In ProptechOS, a digital twin is not only the building as a whole. Every sensor, device, asset, and space is its own digital twin — a structured, RealEstateCore-modeled representation of that physical object with associated metadata, telemetry, and relationships. Onboarding means creating and connecting all of those twins.
Q: Why does the tag list matter so much? BMS systems store sensor data under local, often undocumented naming conventions — sometimes in Swedish, Norwegian, German, or proprietary shorthand. A tag list is the raw input that tells the onboarding agent what exists in the system. Without it, there is nothing to model.
Q: What if my system is not in your connector library? If your system is not among the 90+ existing connectors, ProptechOS will build one. Connector development that previously took a week now takes one to two days. API documentation from the system vendor is required to get started.
Q: How does the agent handle tags it is not confident about? Tags that fall below the automapping confidence threshold are flagged for review. The onboarding agent applies its own reasoning — using known tagging standards for that country or vendor — before deciding whether to map, skip, or escalate. Very low-confidence tags can be reviewed by a human or passed back to the customer for clarification.
Q: How are agent permissions managed during onboarding? The onboarding agent operates on zero standing privileges. Before it begins, it has no access to ProptechOS. It requests only the specific permissions it needs — write access for digital twins in one specific building — points to the Jira task as verification, and receives a temporary, scoped token. When the work is done, those privileges are revoked automatically.
Q: How long does onboarding take now? It depends on portfolio size and system homogeneity. In recent projects: 15 buildings with two BMS systems each, including BIM models, completed in three days. A portfolio of 17 buildings with the same system type completed in three hours. The constraint is almost always access and networking, not the modeling work itself.
