Agentic Workforce Company
Agent Infrastructure · Knowledge

A folder of PDFs isn’t knowledge. So I built one that compiles.

I implemented Karpathy’s LLM Wiki pattern inside my agent vault: drop a source into a read-only folder, a script registers it with a SHA-256 hash, and an AI agent compiles it into interlinked, Wikipedia-style pages that get linted like code. It doesn’t retrieve — it compiles. Raw truth stays immutable; understanding accumulates on top.

Phase 1 · supervisedROLE Architect & ownerPATTERN Karpathy LLM WikiSHIPPED Jun 27, 2026
CategoryAgent infrastructure
PatternKarpathy LLM Wiki (implemented file-native)
Layersraw / wiki / governance
SubstrateObsidian + git vault
StatusPhase 1 · manual, supervised
The Problem

A storage bucket rots.

My vault was becoming a great storage bucket — and raw retrieval finds passages but never builds a connected picture.

Worse, it happily mixes original source truth with model interpretation, and nothing stops a source being silently edited.

I wanted something better: feed real documents into the brain so agents could interrogate a structured, cited knowledge graph instead of brute-force re-reading everything — and so no source could ever be quietly altered or contradicted without a trace.

When I saw Andrej Karpathy’s LLM Wiki pattern, it matched exactly what I was missing: the general pipeline from raw source to queryable understanding.

The Pipeline

Compile, don't retrieve.

Raw truth stays immutable in one layer; the AI compiles interlinked understanding into a second; a governance layer keeps the graph healthy. Ingest one source at a time — register before you compile, link everything both ways, lint early.

Fig.01 — Ingest Pipeline & Three-Layer SeparationLLM Wiki ingest pipeline and three-layer separationAcross the top, a six-stage pipeline flows left to right: a raw source drop, a hashed source record with SHA-256, an AI compile step, interlinked wiki pages, an index and log update, and a lint pass gating a queryable brain. Below, three separated layers are shown: an immutable raw layer, a compiled wiki layer with five page types, and a governance layer holding the schema, ingest workflow, and lint rules.INGEST PIPELINE1 · Dropraw/inbox/immutable2 · RegisterSHA-256 · MIME→ log.md3 · CompileAI agent readsextract / OCR4 · Linkinterlink pagesboth ways5 · Index/Logwiki/index.mddated entry6 · Lint → Querygate commitqueryable brainTHREE STRICTLY-SEPARATED LAYERSRAW · IMMUTABLEraw/source truth · never editedwrong? add a correcting note —never alter the originalmarkdown · pdf · images ·transcripts · media pointersSHA-256 on every sourcecontent-addressed provenanceWIKI · COMPILEDwiki/ — 5 page types• sources• entities• concepts• syntheses• questionsevery page links ≥1 source ·contradictions flagged, not hiddenGOVERNANCE · RULES00-system/llm-wiki-schema.mdllm-wiki-ingest-workflow.mdllm-wiki-lint-rules.mdwiki-register-source.shwiki-lint.sh (exits non-zero)lint the graph like code

Drop a source into the immutable raw/ layer; wiki-register-source.sh hashes it (SHA-256), captures size and MIME, generates a kebab-case source note, and logs a dated entry. An AI agent compiles it into the wiki/ layer’s five page types, linking everything both ways; wiki-lint.sh then checks the graph and exits non-zero to gate the commit. The raw layer is never edited — if a source is wrong, you add a correcting note, you don’t rewrite history.

Scripts

wiki-register-source.shwiki-lint.shidempotent by hashcommit-gating

Schema & Rules

llm-wiki-schema.mdingest-workflow.mdlint-rules.mdkebab-case namingrelative links

Intake

Obsidian Web Clipperraw/inbox/PDF / image / transcript

Compile Engine

AI agent (Claude Code / Codex)extraction / OCRsource-grounded

Substrate

Obsidiangit vaultagentic-brain repo

Deferred (Phase 2+)

file watcherextraction workersauto-compilevector index / GBrain
By the Numbers

A working scaffold with a real first batch.

Verified from the live wiki/ directory and wiki/log.md.

Wiki page types
5
sources · entities · concepts · syntheses · questions
Separated layers
3
raw (immutable) · wiki · governance
Sources registered
~25
first ingest batch · every file hashed & logged
Lint targets
6
broken links · orphans · coverage · drift · contradictions · low-confidence
SHA-256
on every source — content-addressed provenance; exact-duplicate files detected and recorded, not re-ingested
5-phase
automation roadmap documented (manual → watcher → workers → auto-compile → scheduled lint); Phase 1 shipped first
1 public
SEO topic — "How Search Engines Work" — compiled end-to-end as the shareable example; larger batch kept private

The first batch registered ~25 sources: one public SEO topic (“How Search Engines Work”) plus a larger private personal-document batch — each file hashed with SHA-256, extracted where text was available, and logged with its date. The private batch is intentionally not shown; every public visual on this page uses only the neutral SEO example.

Results

What exists and works — and what's honestly Phase 1.

A functioning Phase-1 LLM Wiki: schema, ingest workflow, lint rules, a working registration script, a working lint script — with real ingests completed and logged.

● What exists & works · verified

A working scaffold with a proven ingest.

  • Schema, ingest workflow, and lint rules authored in 00-system/.
  • A working wiki-register-source.sh and wiki-lint.sh.
  • The SEO source compiled into source / entity / concept / synthesis / question pages — including a “How Search Engines Work for SEO” synthesis linking back to its registered source.
  • A larger real-world batch registered, hashed, extracted, and compiled into document maps and explicit open version/OCR questions.
  • Every source logged with its hash and date.
▲ Honest status · Phase 1 by design

Manual, supervised, one source at a time.

  • No daemon, watcher, vector index, or GBrain is wired in yet — those are the documented next phases.
  • It’s a working scaffold with a small proven batch, not a large automated knowledge engine.
  • Compiled material stays draft — it can’t be bulk-promoted into canonical business or client claims without review.

Framed exactly right: "implemented the pattern and ran the first ingests," not "large automated knowledge engine." Phase 1 is manual and supervised on purpose.

wiki/sources/how-search-engines-work.md — generated note (SEO example)
# Sample source note — neutral, non-sensitive example title: How Search Engines Work status: compiled # registered → extracted → compiled → needs-review sha256: 3f9a…c17b # content-addressed provenance mime: text/markdown registered: 2026-06-27 links: [[entities/crawler]] · [[concepts/indexing]] · [[syntheses/how-search-engines-work-for-seo]] # log.md → 2026-06-27 registered how-search-engines-work (sha256 3f9a…c17b)
Before / After

A pile of files vs a linked graph.

A source in the vault
BeforeRaw retrievalfinds passages · mixes source truth with model interpretation · silently editable
AfterCompiled & citedSHA-256 provenance · immutable raw · interlinked pages · linted like code
My Role

I brought the pattern in and sequenced the build.

I judged that Karpathy’s LLM Wiki pattern fit our Obsidian + git + agent-memory foundation, and made the sequencing call: simple file-native version first, raw immutable, GBrain deferred.

I directed the build of the Phase-1 scaffold — the schema, the ingest workflow, the lint rules, and the register/lint scripts — then approved it and greenlit the first ingests. I set the rule that compiled wiki material stays draft and can’t be bulk-promoted into canonical claims without my review. I’m the architect and owner; the agent compiled under my direction. The pattern is Karpathy’s and GBrain is Garry Tan’s — the implementation and integration are mine.

Skills Demonstrated

What this took.

Knowledge-graph / knowledge-base engineeringApplying a published AI pattern (Karpathy LLM Wiki)Source-grounded / provenance-first design (SHA-256)AI-assisted content compilation & governanceShell scripting (register + lint automation)Lint-the-graph-like-code quality disciplinePhased delivery (manual before automation)Document extraction / OCR pipelinesAgent context engineering

Want an operator who turns files into durable understanding?