Integrated Hubspot to larger lead funnel through the API and external databases, and use bayesian machine-learning algorithms to target drip campaigns towards most probable real-estate customer segments by demographics. This was an ETL intensive project because of the need to important contact and property information and enrichments from multiple sources and turn into a cohesive list of de-duplicated leads.
The drip campaigns involved scheduled emails, sms texts, pre-recorded voicemails, post-cards, leading to appointments with regular sales people.
To help a sunglasses wholesale manager their inventory better, Tim Gilbert created a web-interface to automatically ingest PDF reports from the manufacturer, pull in order data from a merchant account API, and create adjustable forecasts of demand for each existing product based on its prior performance or for new products based on comparison to similar past released products, and generate suggested weekly orders to maintain an appropriate level of stock to meet expected demand. He also create an exponentially-weight-moving average in order to rank product sales where even top-sellers might not have sales in a given week. Marco created a dashboard to review and amend suggested order quantities.
Using a custom fuzzy-matching library, Tim Gilbert created a system to find company names regardless of typos, abbreviations, or omitted unimportant words
Featured
This project was lead by Tim Gilbert to create target revenue predictions for any house/condo in the United States if it were used as a full-time short-term-rental investment, taking into account property features, offered services, local entertain factors, market demand and penetration, seasonality, local expenses, etc. We extracted ML features from structured fields and descriptions from 6 years of data across 2 million properties, then filtered down to the 600k most relevant houses for predictions and built deep neural network models to predict yearly revenues, daily rates, long-term lease amounts.
We also created a system for finding nearby comparable properties, based not only on distance and number of bedrooms/bathrooms, but also on key features that affect house price and revenue such as pools, ski-in/out access, waterfront access, and much more.
Can you sign up to use the predictor at runrevr.com.
Example prediction results
Market Comparisons
Tim Gilbert led the effort to use NLP word vectors, entity recognition, and regular expressions to analyze 70 topics across corporate privacy policies from companies across 183 NAICS sectors to find the types of similarities and differences in coverage by sector, and identify which paragraph and sentences in each policy were semantically aligned with each topic.
One challenging part of the project was distinguishing between gathered data that company reserved the right to use internally vs data that it would be sharing/selling to other companies.
Example privacy topics analyzed:
- Kinds of unique identifiers collection
- Biometrics
- Communication logging
- Cookies, cross-origin histories and device tracking
- Interests
- Submitted content ownership
- Employment history
Example external recipients of collected data:
- Law enforcement and government entities
- Advertisers
- Joint ventures
- Parent companies
- 3rd party service providers
- Liquidation purchasers
- Credit bureaus
- Fulfillment agencies
- Brokers
- Analytics Companies