This repository documents the end-to-end architecture thinking process for the given kata scenario prepared by Five Nines Team. This solution is the winner of the Kata Q4 2025: AI-Enabled Architecture
Final presentation recording is available under this link.
- About the Project
- Problem Statement
- Solution Overview
- Risks and Mitigations
- Future Scope
- VIDEO presentation
- References
This project is intended to help MobilityCorp achieve its business goals and address its biggest business challenges by offering the most optimal technology stack and architectural solution. The company's goals are:
- increase sales and revenue,
- expand market coverage,
- improve user experience and satisfaction,
- strengthen its market position.
The project was prepared by the team called 'Five Nines' consisting of:
9 . Oleksandra Tytar
adrs/→ Architecture Decision Recordsrequirements/→ Business & technical requirementshld/→ High Level Design artefacts (diagrams, docs)
Design from scratch (green field architecture) functionality for:
- User Dialogue - enhanced sales via user-personalized companionship, adaptive route/experience/pricing/charging advise, and driving compliance guidance;
- Dynamic Pricing - sales support with more competitive pricing, addresses expansion ambitions and retention goals;
- Demand Forecasting - uses internal and external data (weather, traffic, events) to forecast demand;
- Maintenance Optimization (cost reduction) via:
- Operational efficiency (location/fleet/route supply optimisation, transportation/charge task assignment);
- Load distribution (even rental/usage among vehicles);
- Maintenance prediction (sensor data - e.g. battery);
while ensuring continuous and smooth system operation of the existing core rental functionality.
Our goal is to design a micro-mobility platform for connected scooters, e-bikes, cars, and vans. The value proposition focuses on near-real-time telemetry, fleet state, pricing and demand intelligence, maintenance optimization, and conversational user assistance.
First, we integrated a Trip Copilot AI companion into the MobilityCorp mobile app to recommend routes based on user location and preferences, weather, destination, and riding style, enhancing customer experience and retention.
Second, we used the AI companion to gather customer feedback, which is analyzed to identify improvements and optimal parking bay locations, supporting our expansion strategy.
Next, we focused on vehicle quality and customer satisfaction by ingesting telemetry data into a custom ML model to predict battery replacement timing and prevent ride failures.
Then, we designed an ML-powered system to optimize pricing for customers based on multiple factors.
Lastly, we leveraged historical data and real-time route information to forecast demand and optimize fleet allocation to satisfy it.
Edge and connectivity
- Vehicles use MQTT over cellular
- Clustered MQTT gateway on GKE behind external load balancer
- Gateway authenticates devices, buffers if offline, and publishes telemetry to Pub/Sub topics by region and vehicle type
Streaming and integration
- Pub/Sub as event backbone for telemetry and commands
- Dataflow pipelines for parsing, validation, dedupe, geofencing enrichment, road snapping, and fan-out to storage and APIs
Data and Analytics
- Ingestion: streaming, micro-batch, and bulk batch modes across transactional sources, vehicle telemetry, files, APIs, and reference data
- Persistence: multi-layer lakehouse with raw, bronze, silver, gold, feature store, and curated semantic layer
- Processing: Dataflow (stream/batch), Dataproc/Spark, BigQuery SQL, Vertex AI for feature engineering and ML pipelines; orchestrated with Airflow/Prefect/Dagster
- Analytics Surfaces: governed semantic layer, BI dashboards in Looker, SQL/Notebook workspaces, ML experiment tracking, curated APIs
- Reliability: multi-AZ baseline, optional multi-region with RPO 0–5 min (streaming) and RTO ≈15 min for critical services
- Governance: schema contracts, lineage, RBAC/ABAC, PII tokenization, audit logs ≥400 days
- FinOps & Ops: unit cost tracking, observability (lag, throughput, error rates), structured logging, OpenTelemetry tracing
AI and GenAI
- Vertex AI for model lifecycle and evaluation
- Gemini for map grounded conversational search and ops copilots
- Matching Engine and BigQuery vector search for retrieval and RAG
- Agent Builder for tool calling and orchestration of fleet, maps, and pricing tools
- MLOps for automating full ML model lifecycle
Applications
- Cloud Run microservices for booking, pricing, ops, maintenance
- API Gateway for external and mobile access
- Identity Platform for auth
Geospatial and UX
- Maps Platform for routing, POI, geocoding, traffic, and ETA
- Ops UI and mobile client consume APIs and AI assistants
video link Click the link to watch the video.
- Google Cloud
- Vertex AI
- Gemini (Google AI)
- BigQuery
- MQTT (OASIS Standard v5.0)
- EU Micromobility Regulation Overview (EC Urban Mobility Observatory)
Note: Portions of this document were developed using AI-assisted tools under human supervision. All final content was proofread, reviewed, and approved by the authors who take full responsibility for it.