ai-knowledge-filler 0.5.4

Schema validation pipeline for LLM-generated structured Markdown. CLI · Python API · REST API.

AI Knowledge Filler

Validation pipeline for LLM-generated structured Markdown

---

The Problem

LLMs generate text. You need structured, schema-compliant files.

Without a validation layer, AI-generated Markdown produces:

Error	Raw LLM output	What you need
Enum violation	level: expert	beginner | intermediate | advanced
Domain violation	domain: Technology	domain: system-design
Type mismatch	tags: security	tags: [security, api, auth]
Date format	created: 12-02-2026	created: 2026-02-12

One file? Fixable manually. A hundred files? The schema collapses.

AKF enforces the contract at generation time, not review time.

---

How It Works

Prompt
  → LLM                  (only non-deterministic component)
  → Validation Engine    (binary: VALID or INVALID + typed E-codes)
  → Error Normalizer     (deterministic repair instructions from E-codes)
  → Retry Controller     (max 3 attempts — aborts on identical failure hash)
  → Commit Gate          (atomic write — only VALID output reaches disk)

No silent failures. No partial commits. No guessing.

Retry = ontology signal. When a domain triggers elevated retries, the taxonomy has a boundary problem — not the model. Telemetry captures this.

---

Quick Start

pip install ai-knowledge-filler

export GROQ_API_KEY="gsk_..."   # free tier, fastest

# Generate new file
akf generate "Create a Docker networking guide"
# → Docker_Networking_Guide.md (validated, schema-compliant)

# Enrich existing files — add YAML to files that have none
akf enrich docs/

# Validate an entire directory
akf validate --path docs/

---

AKF Documents Itself

This repo uses AKF to validate its own documentation on every PR.

Setup:

# 1. Define your taxonomy
cat akf.yaml

schema_version: "1.0.0"
vault_path: "./docs"
taxonomy:
  domains:
    - akf-core
    - akf-docs
    - akf-ops
    - akf-spec

# 2. Enrich existing docs — AKF adds frontmatter via LLM
akf enrich docs/ --model groq

# 3. Validate
akf validate --path docs/
# ✅ docs/cli-reference.md
# ✅ docs/user-guide.md
# → Total: 2 | OK: 2 | Errors: 0

CI gate (.github/workflows/validate.yml):

- name: Validate docs/
  run: akf validate --path docs/

Every PR that introduces invalid metadata fails the check. The Validate badge above is AKF validating AKF's own docs.

---

akf enrich

Add YAML frontmatter to existing Markdown files — bulk or single.

akf enrich docs/                    # enrich all .md files
akf enrich docs/ --dry-run          # preview only, no writes
akf enrich docs/ --force            # overwrite valid frontmatter
akf enrich docs/ --output enriched/ # copy to output dir

File state	Default	--force
No frontmatter	Generate + validate + write	Same
Incomplete frontmatter	Fill missing fields only	Regenerate all
Valid frontmatter	Skip	Regenerate all
Empty file	Skip with warning	Skip

Enrich runs through the same validation pipeline as generate — retry loop, commit gate, telemetry.

---

Python API

from akf import Pipeline

pipeline = Pipeline(output="./vault/", model="groq")

# Generate new file
result = pipeline.generate("Create API rate limiting guide")
print(result.success)        # True
print(result.path)           # PosixPath('vault/API_Rate_Limiting_Guide.md')
print(result.attempts)       # 1 (retried if schema violation)

# Enrich existing file
result = pipeline.enrich("docs/old-note.md")
print(result.status)         # "enriched" | "skipped" | "failed"

# Enrich directory
results = pipeline.enrich_dir("docs/")

# Batch generate
results = pipeline.batch_generate([
    "Docker deployment best practices",
    "Kubernetes security hardening",
    "API authentication strategies",
])

# Validate
v = pipeline.validate("vault/my_file.md")
print(v.valid, v.errors)

---

REST API

akf serve --port 8000

curl -X POST http://localhost:8000/v1/generate \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Create Docker security checklist", "model": "groq"}'

curl -X POST http://localhost:8000/v1/batch \
  -H "Content-Type: application/json" \
  -d '{"prompts": ["Docker guide", "Kubernetes guide"]}'

curl -X POST http://localhost:8000/v1/validate \
  -H "Content-Type: application/json" \
  -d '{"content": "---\ntitle: Test\n..."}'

Endpoints: POST /v1/generate · POST /v1/enrich · POST /v1/validate · POST /v1/batch · GET /v1/models · GET /health

Swagger UI: http://localhost:8000/docs

---

What Every Committed File Guarantees

• Required fields: title, type, domain, level, status, tags, created, updated
• Valid enums: type, level, status from controlled sets
• Domain from configured taxonomy (akf.yaml) — not hardcoded
• ISO 8601 dates with created ≤ updated
• tags as array (≥3), title as string

Error Codes

Code	Field	Meaning
E001	type / level / status	Invalid enum value
E002	any	Required field missing
E003	created / updated	Date not ISO 8601
E004	title / tags	Type mismatch
E005	frontmatter	General schema violation
E006	domain	Not in taxonomy
E007	created / updated	created > updated

---

Configuration

# akf.yaml
schema_version: "1.0.0"
vault_path: "./vault"

taxonomy:
  domains:
    - ai-system
    - api-design
    - devops
    - security
    - system-design
    # add your own

enums:
  type: [concept, guide, reference, checklist, project, roadmap, template, audit]
  level: [beginner, intermediate, advanced]
  status: [draft, active, completed, archived]

akf init          # creates akf.yaml in current directory
akf init --force  # overwrite existing

---

CLI Reference

# Generate
akf generate "prompt" [--model groq|claude|gemini|gpt4|ollama] [--output PATH]

# Enrich
akf enrich PATH [--dry-run] [--force] [--model MODEL] [--output DIR]

# Validate
akf validate [--file FILE] [--path PATH] [--strict]

# Server
akf serve [--host HOST] [--port PORT]

# Models / Init
akf models
akf init [--path DIR] [--force]

---

Model Selection

Model	Key	Speed	Cost	Notes
Groq	GROQ_API_KEY	⚡	Free tier	Recommended for CI, high volume
Claude	ANTHROPIC_API_KEY	Medium	$$$	Technical docs, architecture
Gemini	GOOGLE_API_KEY	Fast	$	Quick drafts
GPT-4	OPENAI_API_KEY	Medium	$$	General purpose
Grok	XAI_API_KEY	Fast	$$	General purpose
Ollama	—	Fast	Free	Local / offline / private

Auto-selection order: Groq → Grok → Claude → Gemini → GPT-4 → Ollama.

---

Telemetry

Each generation appends a structured event to telemetry/events.jsonl:

{
  "generation_id": "uuid-v4",
  "document_id": "abc123",
  "schema_version": "1.0.0",
  "attempt": 1,
  "converged": true,
  "timestamp": "2026-02-27T14:22:01Z",
  "model": "groq",
  "temperature": 0
}

Append-only. Never influences the pipeline at runtime.

---

Security

export AKF_API_KEY="your-secret"          # optional — unset = dev mode
export AKF_CORS_ORIGINS="https://app.com"

Rate limits: POST /v1/generate 10/min · POST /v1/validate 30/min · POST /v1/batch 3/min

---

Quality

• 542 tests, 93.74% coverage
• CI green on Python 3.10 / 3.11 / 3.12
• Type hints: 100%
• Pylint: 9.55/10

---

Roadmap

Shipped

• akf generate, akf enrich, akf validate, akf serve, akf init
• Validation pipeline — E001–E007, retry loop, commit gate
• Telemetry — append-only JSONL, ontology friction metrics
• Config layer — external akf.yaml, no code changes for taxonomy
• Pipeline API — from akf import Pipeline
• REST API — FastAPI, rate limiting, optional auth
• Self-documentation — AKF validates its own docs/ on every PR

Planned

• akf generate --batch topics.txt
• Graph extraction layer
• n8n / Make integration templates

---

Documentation

• Architecture — Module map, data flow, extension points
• CLI Reference — All commands, flags, env vars, exit codes
• User Guide — Quickstart, enrich workflow, CI integration
• Contributing — Dev setup, adding providers

---

License

MIT — Free for commercial and personal use.

---

PyPI: https://pypi.org/project/ai-knowledge-filler/ | Version: 0.5.0