The heading-path prefix on chunks is such a small change for the improvement it gives. We retrofitted it last month and recall@5 went from 0.68 to 0.81 with no other modifications. Did you find a good approach for table...
on “Building RAG Pipelines That Survive Production” · 3 days ago →Priya Raman
Senior/u/priya
Platform engineer. Kubernetes, Terraform, and too much YAML.
- Articles
- 0
- Comments
- 4
- Reputation
- 310
Recent comments
That is a much clearer way to frame it than the docs manage. Our spares turnaround is about a week, so we should probably plan for five.
on “Proxmox VE Cluster Design That Survives Failure” · 4 days ago →Question on the Ceph sizing — you say five nodes minimum for production. We are running three with `size=3, min_size=2` and it has been fine for about a year. Is the recommendation about recovery headroom specifically?
on “Proxmox VE Cluster Design That Survives Failure” · 6 days ago →Scenario four in the failure testing section — session up, traffic blackholed — is what got us last year. BGP was perfectly healthy on both links for 40 minutes while half our traffic went nowhere. External reachability...
on “BGP Multihoming for the Enterprise Edge” · 1 week ago →