The Intelligence Behind Everything We Build.

Our matching system analyses over 40 data signals per job request — including job complexity, distance, equipment requirements, historical performance, availability windows, and customer preferences — to rank service providers in real time.

At its core, the engine uses a gradient-boosted ranking model (XGBoost) combined with collaborative filtering trained on completed job outcomes. The system continuously retrains on new data, improving precision with every interaction on the platform.

Static pricing in service marketplaces creates two problems: providers under-price in slow periods and over-price in busy ones; customers never know if they are paying a fair rate. Our dynamic pricing engine solves both by modelling real-time supply and demand.

We operate a sealed-bid composite auction where each quote is scored on a weighted composite (70% quality signals, 30% price competitiveness). A LightGBM demand forecasting model predicts market conditions 7 days forward to help providers price more accurately.

Marketplaces are adversarial environments. Fake providers, inflated reviews, identity fraud, and price manipulation erode user trust rapidly if left unchecked. Our fraud detection system is a 4-layer verification pipeline designed to catch bad actors before they interact with customers.

Layer 1: ABN validation against the Australian Business Register. Layer 2: ASIC registration check. Layer 3: Social proof triangulation (reviews, references, external signals). Layer 4: Behavioural anomaly detection using a real-time rule engine and ML classifier. Each provider receives a Trust Score from 0–100; a threshold of ≥72 is required to receive jobs.

Moving jobs are complex: customers describe moves in natural language ("2-bedroom unit, 3rd floor, no lift, piano, 12km away"), and extracting structured job parameters from that description is a hard NLP problem. Our NLP layer transforms free-text inputs into structured job objects in real time.

We use GPT-4o with structured output schemas for primary extraction, backed by a fine-tuned classification model for intent classification and urgency detection. Sentiment analysis on reviews and in-app messages flags at-risk interactions for human review.

Verifying the identity of service providers and understanding the physical scope of a job are two hard problems that computer vision solves efficiently. Our CV pipeline handles vehicle identification, document verification, and volume estimation from customer-uploaded photos.

Provider vehicle photos are verified against registration records using a fine-tuned Vision Transformer (ViT). Document authenticity is confirmed via GPT-4o Vision. Customers can upload room photos for volume estimation — our model produces ±15% volume accuracy, enabling instant indicative quotes before a human provider reviews the job.