Every Droppy delivery is powered by a stack of specialised ML models that match drivers, price orders, predict ETAs, detect fraud, and optimise routes — all in under 200 ms.
0.0%
ETA Accuracy
0.0%
Match Success Rate
0%
Cost Reduction (AI routing)
0%
Fuel Saved per Trip
Click any model to explore how it works under the hood.
When an order is placed, the matching engine scores every available driver within a dynamic radius using a multi-factor weighted algorithm. The decision is made in under 80 ms.
Proximity Score
Haversine distance + road network ETA
Driver Rating
Rolling 30-day EWMA of ★ scores
Route Efficiency
Post-delivery return-to-hotzone score
Acceptance Rate
Recent acceptance behaviour
Vehicle Suitability
Parcel type × vehicle capacity match
Model: Gradient-boosted ranking (LightGBM)
Latency: p99 < 80 ms
Retraining: Daily on last 30 days of trips
0.8 km
ETA 6 min
1.2 km
ETA 9 min
1.5 km
ETA 11 min
2.1 km
ETA 15 min
Matched James M. — highest composite score across proximity, rating, and route efficiency.
All services communicate via Kafka event streams. No synchronous inter-service calls in the critical path.
Order, Driver, Payment, and AI domains each own their data store. No cross-domain DB joins.
All services run as Docker containers on ECS Fargate or EKS. Zero server management overhead.
A clear migration path from Firebase-first seed stage to a fully distributed multi-region AWS architecture as order volume scales.
Infrastructure
AWS Primary
Estimated Cost
$4,200/mo
Services Used
Full migration to AWS. ECS Fargate for containers, Aurora for ACID transactions, MSK for event streaming. Auto-scaling handles 10× bursts.
Seed
0–1K/day
$120/mo
Growth
1K–10K/day
$800/mo
Scale
10K–100K/day
$4,200/mo
Enterprise
100K+/day
Custom
Every layer of the Droppy AI platform, detailed.
Access all five AI engines through the same REST API. No ML expertise required — just send an order and let the models handle the rest.
© 2026 Droppy Technologies Ltd. · droppyglobal.com