Yoann Abriel
fr

All projects

2024-2026

Cloud Benchmark Dashboard

From a manual Excel benchmark and watch (September 2024) to a production pricing-intelligence platform: automated collectors plugged into provider APIs, 7 European cloud providers and hyperscalers, 12 LLM providers. Supported a repricing review of the offer in spring 2026, then handed over to the team.

Cloud Benchmark Dashboard 1:1 matching module: four price charts by VM size and provider, with a detail tooltip

fig. 01 · The 1:1 matching module: price per physical core, size by size, provider by provider

Cloud Benchmark Dashboard is the industrialization of work that started by hand: in September 2024, competitive intelligence and pricing benchmarks lived in Excel. The production platform is where that ended up. Collectors automatically ingest data through provider APIs (Cloud Mercato only as a complement), across a scope covering 7 European cloud providers and hyperscalers (Cloud Avenue, Scaleway, Outscale, OVH, AWS, Azure, GCP) and 12 LLM-as-a-Service providers. The application features 8 analysis modules: VM Instances, Storage, GPU, Software Licensing, Bare Metal, Dedicated Clusters, LLM-as-a-Service, and OVH comparison. Data is managed via PocketBase with Clerk authentication and Sentry monitoring. In spring 2026, the platform supported a repricing review of the offer (pricing arbitrations). It was taken over by the program management team to continue after my departure: the best proof of its usefulness.

2,156 VM instances compared

Cloud Benchmark Dashboard VM instances table: 2,156 instances from seven providers with vCPU, RAM, price per hour, per vCPU, per GB of RAM and per month, per-column filters

fig. 02 · The VM Instances module: 2,156 instances from 7 providers, sortable and filterable column by column

The VM Instances module lines up the offers of 7 European cloud providers and hyperscalers (Cloud Avenue, Scaleway, Outscale, OVH, AWS, Azure, GCP) in one TanStack table: sorting, per-column filters, comparison and CSV export.

1:1 matching and overcommit

The 1:1 matching module with the overcommit map, documenting each offer's vCPU per physical core ratio

fig. 03 · The overcommit map: the vCPU to physical core ratio documented for each offer

The 1:1 matching module compares VM sizes at the price per physical core, and the overcommit map documents each offer's vCPU per physical core ratio.

Methodology in plain sight

Overcommit sources and methodology panel: methodology column and confidence levels on the left, Cloud Avenue Eco, Standard and High Performance cards with mechanism, ratio, technical explanation and sources

fig. 04 · Overcommit sources and methodology: mechanism, ratio, explanation and confidence level per class

Each instance class carries its allocation mechanism, its overcommit ratio, a technical explanation, its sources and a confidence level (official, strong indicator, inferred, rumor). Normalizing data from 7 providers with different formats stays readable and open to challenge.

Challenges

  • Industrializing a pricing watch that first lived in a hand-maintained Excel
  • Aggregating and normalizing pricing data from 7+ cloud providers with different formats
  • Interactive visualization of large datasets with advanced filtering and sorting
  • Role-based authentication to control access to sensitive data
  • Preparing a clean handover so the tool outlives my departure

Solutions

  • Automated ingestion collectors plugged into provider APIs, with Cloud Mercato as a complement
  • Next.js 16 architecture with TanStack Table for performant filtering and sorting
  • Recharts for interactive and comparative visualizations
  • Clerk for authentication and role management
  • PocketBase backend for data persistence with Sentry for monitoring

Results

  • Supported a repricing review of the offer in spring 2026 (pricing arbitrations)
  • Successful handover: taken over by the program management team to continue after my departure
  • 8 analysis modules covering all cloud offerings
  • Comparison of 7 European cloud providers and hyperscalers, and 12 LLM providers
  • PDF and Excel export of benchmark reports, light/dark theme, command palette

Technologies

Next.js 16 · React 19 · TypeScript · PocketBase · Clerk · Recharts · TanStack Table · Tailwind CSS · Sentry