[lbliii]

Summary

Remove profile/README.md so the NVIDIA-NeMo GitHub organization landing page uses GitHub's native organization layout (org bio, top-level docs link, pinned flagship repositories, and filterable repository catalog) rather than a manually maintained org-level Markdown file.

---

Why Remove the Organization Profile README?

Maintaining a centralized, multi-column technical matrix in NVIDIA-NeMo/.github has created a structural maintenance trap: it is chronically out of date, suffers from multi-owner consensus deadlock, and diverges from standard practice across both NVIDIA and peer AI organizations.

Below is the empirical data from a live audit of 25 peer GitHub organizations and the commit/PR history of NVIDIA-NeMo/.github.

---

1. Industry & NVIDIA Precedent: 80% of Peer Orgs Do Not Use a Profile README

We audited 25 peer AI labs, ML infrastructure ecosystems, NVIDIA organizations, and developer platforms via the GitHub API (GET /repos/:org/.github/contents/profile/README.md).

• 20 of 25 (80%) have no profile/README.md at all.
• 0 of 7 major AI model/research orgs (huggingface, openai, anthropics, google-deepmind, meta-llama, deepseek-ai, mistralai) use an org profile README.
• Core ML framework & infra orgs (pytorch, vllm-project, ray-project, sgl-project, jax-ml) omit profile/README.md and rely on pinned repositories and canonical documentation sites.
• Sibling NVIDIA orgs (NVIDIA, rapidsai, nv-morpheus) have .github repositories for shared CI/templates, and none of them publish a profile/README.md.
• The 20% that do keep it tiny and evergreen: Among the 5 orgs that have one, lightning-ai is 2 sentences (199 B) and microsoft is a static pointer to their open-source portal and Code of Conduct (1,135 B). At 8,535 B, NVIDIA-NeMo has the largest and most complex org profile README in the entire cohort (2.7x larger than the next largest).

Full 25-Organization Comparative Benchmark Table (Oct 2026)

GitHub Organization	Cohort	profile/README.md?	Size (Bytes)	Landing Page Strategy
huggingface	AI Lab / Model Org	No	0 B	Pinned repos, org bio, and docs link
openai	AI Lab / Model Org	No	0 B	Pinned flagship SDKs and cookbooks
anthropics	AI Lab / Model Org	No	0 B	Pinned SDKs and quickstarts
google-deepmind	AI Lab / Model Org	No	0 B	Empty .github repo; pinned repos
meta-llama	AI Lab / Model Org	No	0 B	Pinned repositories only
deepseek-ai	AI Lab / Model Org	No	0 B	Pinned model repositories
mistralai	AI Lab / Model Org	No	0 B	Pinned inference and model repos
pytorch	AI Infra / Framework	No	0 B	Clean repository list and pinned core libraries
vllm-project	AI Infra / Framework	No	0 B	Pinned vllm repo and docs.vllm.ai link
ray-project	AI Infra / Framework	No	0 B	.github repo holds only workflows and templates
sgl-project	AI Infra / Framework	No	0 B	Pinned repositories
jax-ml	AI Infra / Framework	No	0 B	Empty .github repo; pinned repositories
NVIDIA	NVIDIA GitHub Org	No	0 B	Org header metadata + pinned repos
rapidsai	NVIDIA GitHub Org	No	0 B	.github repo used strictly for CI/shared config
nv-morpheus	NVIDIA GitHub Org	No	0 B	.github repo used for shared org config
astral-sh	Developer Platform	No	0 B	Pinned uv, ruff, ty repositories
vercel	Developer Platform	No	0 B	Pinned next.js, ai, turborepo
supabase	Developer Platform	No	0 B	Pinned core repositories
cloudflare	Developer Platform	No	0 B	Pinned workers and SDK repositories
google	Developer Platform	No	0 B	.github repo holds only org config
lightning-ai	AI Infra / Framework	Yes	199 B	2-sentence evergreen tagline; zero repo tables
microsoft	Developer Platform	Yes	1,135 B	Evergreen pointer to opensource.microsoft.com & CoC
open-mmlab	AI Infra / Framework	Yes	1,482 B	Short overview + links to docs/community
langchain-ai	AI Infra / Framework	Yes	2,986 B	High-level bullet list + docs/forum links (no matrix)
triton-inference-server	NVIDIA GitHub Org	Yes	3,180 B	Static backend summary + pointer to docs.nvidia.com
NVIDIA-NeMo	NVIDIA GitHub Org	Yes	8,535 B	7-column manual table across 17 repos, hardcoded commit SHAs, WIP diagram

---

2. Chronic Staleness & Drift in NVIDIA-NeMo

Because NVIDIA-NeMo is a fast-moving modular organization with dozens of independent repositories, a manually curated feature matrix in .github falls out of date almost immediately:

• 43% of active public repositories are missing: NVIDIA-NeMo currently has 30 active, non-archived public repositories (excluding .github), but the table in profile/README.md only lists 17. Thirteen active public repositories are completely absent from the table:
  • Lens, UserSim, DataDesignerPlugins, ProRL-Agent-Server, SDG-PGMs, agent-hardener, eval-author, nemo-compass, nemo-helix, FW-CI-templates, labs-OO-Agents, labs-Voice-Agent, and labs-molt.
• Brittle, hardcoded deep links: The Getting Started table links to hardcoded commit SHAs in NVIDIA-NeMo/RL (blob/fa379fffbc9c5580301fa748dbba269c7d90f883/examples/configs/dpo.yaml#L99 and grpo_math_1B_megatron.yaml#L79) rather than stable documentation.
• Permanent "WIP" state & typos on the public landing page:
  • Section heading reads ### Diagram Ilustration of Repos under NeMo Framework (WIP) (with both the spelling error "Ilustration" and the (WIP) label sitting live on our primary GitHub landing page since August 2025).
  • The copyright header still reads 2024-2025.

---

3. Multi-Owner Consensus Deadlock & Maintenance Tax

• 94% of repository PR history is README churn: Out of 36 pull requests ever opened in NVIDIA-NeMo/.github, 34 PRs (94%) were solely to edit the README, add a row, or fix broken Markdown table syntax and typos (e.g., back-to-back table/typo fixes in #20, #21, #22, #26, #27). Only 2 PRs in the history of the repo (#1 for CODEOWNERS and #11 for ISSUE_TEMPLATE) were for actual org-wide GitHub configuration.
• Tragedy of the commons in review: .github/CODEOWNERS assigns profile/ to 4 reviewers, while the content attempts to represent 30 independent repositories and sub-teams.
  • When an individual repo changes its scope, container strategy, or docs URL, its maintainers have to open a second PR here and wait for cross-org approval.
  • Conversely, broader structural updates to the README stall indefinitely because it is nearly impossible to reach consensus across all repo owners on how every project is weighted and described (see #29, open since April 2026, and #32, open since May 2026).

---

4. What Replaces It? (Why GitHub's Native UI Works Better)

Removing profile/README.md does not leave the organization page blank—it promotes GitHub's native discovery surfaces above the fold:

1. Pinned Repositories Above the Fold: Right now, visitors to github.com/NVIDIA-NeMo must scroll past an 8.5 KB wall of tables and a large PNG diagram before seeing a single repository card. Removing the profile README brings Pinned Repositories (Automodel, RL, Megatron-Bridge, Curator, Guardrails, Evaluator, etc.) directly to the top of the page, with live star counts, descriptions, and languages automatically synced from each repo.
2. Decentralized Autonomy: Each repository team owns its own repo description, topics, and README without needing approval in NVIDIA-NeMo/.github.
3. Canonical Docs for Deep Comparisons: Multi-repo architectural matrices and getting-started guides belong in our versioned documentation site (docs.nvidia.com/nemo), which is linked directly from the GitHub organization header.

[lbliii]

@snowmanwwg or @bernardwin could I get your approval?

[bernardwin]

Agreed.