Overview
Rohit G (GitHub: rohitg00) is the author of the LLM Wiki v2 gist, a significant production-oriented extension of Karpathy’s original LLM Wiki pattern. He also created agentmemory (20K+ Stars), a persistent memory engine for AI coding agents, and pro-workflow (2.8k⭐, 41 skills, self-correcting memory + knowledge plane on SQLite).
Key Facts
- Author of LLM Wiki v2 — extends Karpathy’s pattern with lifecycle management, knowledge graphs, and automation
- Creator of agentmemory (20K+ Stars on GitHub) — persistent memory engine for AI agents
- Author of pro-workflow (2.8k⭐, 278 forks, 86 commits) — self-correcting memory + persistent FTS5 wikis + budget-capped auto-research loop on a single SQLite store; 41 skills, 8 agents, 23 commands, 37 hooks across 24 events; cross-agent via
skills add - Built on the iii-engine (iii-hq/iii) framework
- Reported 95.2% on LongMemEval-S using BM25 + vector + knowledge graph with RRF fusion
- His gist has 1,554+ stars and 213+ forks on GitHub
Relationships
- “extends” llm-wiki — Extended the original pattern with production lessons
- “introduces” memory-lifecycle — Introduced confidence scoring, consolidation tiers, and forgetting curves
- “advocates for” knowledge-graph — Advocated for typed relationships and graph traversal over flat wikilinks
- “operates with” claude-code — Primary tool for operating an LLM Wiki in practice
- “created” pro-workflow — Author of pro-workflow, the professional workflow layer that compounds corrections over 50+ sessions
Additional Context (2026-09-02 — pro-workflow)
pro-workflow consolida as ideias de LLM Wiki v2 e agentmemory em um produto operacional: uma SQLite store única sob todas as sessões com FTS5 + wikis persistentes + auto-research com budget. Para este vault, é a referência mais direta de como evoluir raw/→wiki/ para FTS5 shadow index e auto-injeção por tópico.
Implications: o autor deixa de ser só “teórico de LLM Wiki” e passa a ser fornecedor de um stack completo (memory + knowledge plane + quality gates), o que aumenta a relevância de acompanhar seus releases para decisões de arquitetura deste vault.
Implications
Rohit G’s work represents the most significant single-extension of the LLM Wiki pattern beyond Karpathy’s original. His production emphasis (what breaks at scale, automation, quality controls) directly informs this vault’s next improvement priorities. The memory lifecycle and hybrid search concepts are the two pieces most immediately applicable. Com pro-workflow, a ênfase migra de “pattern” para “produto” — memória que compõe + knowledge plane operacional.
External Links
Sources
^[raw/articles/llm-wiki-v2-rohitg00-2026.md] ^[raw/articles/raindrop-second-brain-collection-2026.md]