Research
>>> We aim to offer data driven consulting applications in the real-estate industry.
- Data: non-anecdotal, manually collected input;
- Model: Cutting-edge anomaly detection based on neurobiology (HTM);
- Scope: Multi-instance, multi-vector, multi-modal test environment;
- Hardware: Proprietary data center and power supply.
Input
We track consumer behaviour using our proprietary online platform, eDezvoltator.ro - an aggregator that indexes all the residential projects in Romania and provides its users advanced filtering and selection options. We measure both active and passive user interaction and attribute it to specific features previously collected (field data) from the real estate project. This method allows us to measure consumer trends and interest relative to specific offer characteristics without the need of anecdotal data.
For each residential project we collect, classify and verify more than 400 variables, both from the overall building and from the individual units. We manually standardize the data, especially that related to the locative units - such as the floor plan, thus gaining to an unprecedented depth of information.
Model
Our mission is to go beyond current machine learning approaches and replicate as much as possible our natural neuro-biological functions and circuitry, essentially building a machine that works like the human brain. We can then holistically use this replica to solve real-life business problems as opposite to using specific machine learning algorithms to solve isolated problems.
We vectorize the data using Natural Language Processing (NLP) derived encoding methods such as Bidirectional Encoder Representations from Transformers (BERT) and multi-modal technologies inspired by Google`s Multitask Unified Model (MUM) approach.
Core tasks such as classification and anomaly detection are handled by standard machine learning models, augmented with Numenta`s Hierarchical Temporal Memory (HTM) technology.
The market simulation environment is based on proprietary data models and is populated with real-data modeled virtual agents that execute hypothesis testing using variations of Generative Adversarial Networks (GAN) like BiGAN.
Inference quality control is done with real-life testing through our lead-generation platform.
Output
Our system is used test variations of real estate development project using real-life data collected from the market. This is achieved by simulating the supply (residential developments), the demand (consumers) and the context (macro-economy, "black swan" events, geopolitics etc.) in a variety of situations raging from stable conditions and moderate variations to extreme conditions and versatile variations.
Being an inovative project, it is high risk/low-yield, especially at the initialisation phase. Therefore, we want to avoid as much as possible external financiar dependency - we believe that such dependency will limit our decision capacity and orient our project away from science and toward financial gain. Thus, we embedded the comercial component (lead-generator) in the research project such that the project is both auto-sustaining and auto-scalable, without influencing it`s focus. The lead-generation is a core component of the research project, being the main consumer data source and cannot be separated. We will accept and pursue scientific grant type funding.
Hardware
Currently we rent cloud based computing power to do basic research tasks. This is not only expensive, but it is also limiting due to queues and hardware failure. We encounter this type of constrains from all the main suppliers - Google, Amazon and Microsoft.
In order to overcome this challenge, we aim to build our own computing center near the city of Sibiu, where we bought a plot of 10 hectares (24 acres). Situated on a sunny flat field, through which passes the Porumbacu river, our data center will have to be able to extract power both from solar (photovoltaic panels) and hidro (Turbulent Vortex Turbine) sources. The cooling will also be augmented by the river.
Timeline
2017 - 2020 - Lead generator and market research
- Market research - operated realtor office near Bucharest for 6 months
- Funded eDezvoltator.ro - lead generator, data collector and initial income source
- Optimised operations - SEO based marketing, low-wage data input-based workforce
2020 - 2025 - Funding and data collection
- Break even - monetization of lead generator
- Supply data - identifying and auditing all residential projects
- Demand data - scale lead-generator with marketing campaign
2023 - 2025 - Model testing and refinement
- Strategic partnership - identify and collaborate with hedge fund or large-scale developer
- Agent spawning - model context and populate with primitive agents
- Model refinement - increase agent and context complexity
2025 - 2030 - Model scaling and data center
- Adversarial agents - Switch to GAN or similar technology
- Live testing - live output testing using lead-generator
- Data center - power supply and computing compound
contact
If you find our research interesting or wish to partner with us, please reach out:
Razvan Radescu - Project Manager
+40 724 632 719