Training Frameworks and Tooling
Collection
QLoRA training, corpus curation, deterministic pipelines, and WORM-sealed fine-tuning infrastructure. β’ 11 items β’ Updated
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Android app (Kotlin + Jetpack Compose) for OCR'ing Ahmad's handwritten notebooks into THE BOOK digital corpus and feeding the results to the sovereign agent stack.
snapkitty-shell :3700 β /ocr/ingest β triggers THE_BOOK workflowROBOB :4600 β /ocr/book_page β Brain analysis via BedrockABZU IDE :4000 β /api/book_page β live feed in Phoenix LiveView./gradlew installDebugIn AgentBridge.kt, replace 10.0.2.2 with your machine's LAN IP:
private val SHELL_URL = "http://192.168.1.X:3700/ocr/ingest"
Use consistent tags so THE BOOK assembles correctly:
THE_BOOK_CH1 β Master Toltec Mitote
THE_BOOK_CH2 β A Patient Mind
THE_BOOK_CH3 β Past Present Future
THE_BOOK_CH4 β Mastery of Intelligence
THE_BOOK_CH5 β Sacred Taxonomy
THE_BOOK_CH6 β Mothers Poem (poetry only β no code)
THE_BOOK_CH7 β Soul to Code
THE_BOOK_CH8 β Agent Illumination Protocol
THE_BOOK_NEW β New pages not yet assigned to a chapter
Android App (Kotlin)
βββ OcrProcessor.kt β ML Kit text recognition, SHA-256 WORM hash
βββ AgentBridge.kt β HTTP push to shell/ROBOB/ABZU (graceful degradation)
βββ MainActivity.kt β Compose UI: gallery picker + camera + results list
Sovereign Stack (receives)
βββ snapkitty-shell β indexes page into THE_BOOK corpus
βββ ROBOB β Brain (Claude Sonnet 4.6 Bedrock) analyzes page
βββ ABZU IDE β live feed in GITLAB tab (or new BOOK tab)
Each page is hashed as: SHA-256(tag:pageIndex:timestamp:ocrText)
The first 16 hex chars become the worm_hash displayed in the app and sent to agents.
⬑ Ω ⺠Ψ ΠΠΣ Φ α SnapKitty Collective · 2026