Assessment Engine
Handles real-time scoring, branching logic, and instant result delivery. Processes up to 600 answers per second during peak national events without queuing.
Node.js Redis WebSockets
Technologies & Stack
14 core technologies actively running on the platform right now
How the platform is built
Posametul has been running on a carefully selected stack since 2021 — one chosen for reliability under load, not for novelty. Each tool earns its place by solving a specific problem at scale.
What runs underneath
Six layers, one coherent system
Each part of the stack was chosen to handle a specific bottleneck — from delivering 8,000 simultaneous quiz sessions without latency to keeping question banks version-controlled and auditable.
Content Storage
Question banks, media assets, and learner records sit in a structured store with full version history. Every change is logged — no silent overwrites.
PostgreSQL S3-compatible pgvector
Auth & Identity
Role-based access covers 9 distinct permission levels — from a guest preview to a regional coordinator editing live tests. Sessions expire on inactivity within 12 minutes.
OAuth 2.0 JWT RBAC
Analytics Pipeline
Aggregates answer patterns, time-on-question data, and cohort comparisons. Reports refresh every 90 seconds during live events and every 6 hours for trend dashboards.
ClickHouse Kafka Grafana
Frontend Delivery
The quiz interface loads fully in under 1.4 seconds on a 4G connection. Static assets are edge-cached across 12 Canadian PoPs for consistent performance across provinces.
React Vite CDN edge
Infrastructure
Containers orchestrate the workload, auto-scaling between 3 and 48 instances based on active session count. Deployments roll out without downtime — tested on 137 releases to date.
Kubernetes Docker Terraform
Performance by the numbers
Before and after the 2023 rebuild
The platform was re-architected in Q3 2023 after load tests at 4,200 concurrent users revealed bottlenecks in the scoring layer. These numbers reflect what changed across 4 key dimensions.
Results vary across quiz types and session sizes — live competitive events produce the most distinct contrast with earlier benchmarks.
38ms
Median answer latency, down from 410ms
99.7%
Uptime across rolling 365-day window
Decisions and reasoning
What shaped the technical choices
Remote access from 11 provinces and territories puts real constraints on what "fast enough" actually means. The stack reflects those constraints more than it reflects any preference for particular tools.
Constraints that drove architecture
Canada's geography means a participant in Fort St. John and one in Halifax both expect a fair, low-latency quiz experience. Edge caching and regional routing solve that problem more reliably than raw server power.
- Question delivery routed through 12 edge nodes across CA
- Offline-tolerant quiz mode for participants with unstable connections
- All personal data stored on Canadian infrastructure, compliant with PIPEDA
- Bilingual content pipeline — EN/FR without duplicating database records
- Accessibility: WCAG AAA across the quiz interface since March 2024
Where the roadmap is heading
Three projects are in active development for 2025: an adaptive difficulty engine that adjusts question selection based on a participant's answer pattern within the same session, a public API for institutional partners to embed quizzes in their own LMS, and a reworked result dashboard that replaces score percentages with skill maps.
137 releases deployed without downtime
12 edge nodes serving Canadian regions
8 min median incident recovery time
Live monitoring during a national quiz event — 6,300 participants active simultaneously
Built to be questioned
Every architectural decision is documented in an internal decision log. When something changes — a library is swapped, a service is deprecated — the reasoning is written down, not just the outcome. The platform earns trust by being transparent about what it runs and why.