A busy sprint with 75 tickets in flight. The centre of gravity is Project Ratings — a redesigned report, historical run capture, a new validation framework, and an extended rating summary — alongside the Isometric registry integration, the ongoing website refresh, and improvements to chat and research. Highlights:
Project Ratings — new report design
A redesigned Project Ratings report is in build — new layout, refreshed content, and an updated generation flow across the entire report.
Project Ratings — historical run capture
Capturing the underlying rating feature data from recent pipeline runs, so movements in a project's rating can be traced back to exactly what changed between runs.
Project Ratings — validation framework
A new framework that automatically validates rating outputs after each pipeline run — an important step-change in the quality assurance around our ratings.
Project Ratings — extended overall rating summary
The overall rating summary is being extended to give a fuller, clearer picture of how a project's rating comes together.
Isometric registry integration
Analysis and build of the data pipeline for the Isometric registry is underway, including surfacing Isometric data in Carbon Insight reports and the groundwork for rating these projects.
Registry document sync
Keeping project documents automatically in sync with their registries — expired and last-seen document tracking is rolling out across further registries, and registry-sourced documents now take precedence over duplicate uploads.
Website refresh
Build of the first CarbonAI Chat sub-solution page is in progress, with designs for the About Us and Pricing pages taking shape and the country-level data dashboard moving to the new frontend stack.
Chat & research improvements
Research sessions are gaining follow-up context, so successive questions build on earlier answers, alongside work to improve the handling of large document uploads in chat.
Platform & deployment
Deployment pipelines are being consolidated and database usage patterns analysed to right-size autoscaling.