Venkata Peetla

MOD · Model Plane

Model Plane taxonomy

Adaptation methods, task types, classical ML lane, and how ModelForge + DomainForge compose — every row labeled Demo vs Strict / receipt vs educational.

/technical-reviewModelForge deep diveDomainForge deep dive

Compose topology

Model Plane — buy vs RAG vs PEFT · CUDA receipts · gateway enforce+recordArchitecture
TRIAGE & FACTSADAPT & SERVEROUTE & RECORDBuy vs RAG vs PEFTEnterprise RAGfactsDomainForgePEFT trainSLM bake-offPEFT receiptCUDA T4vLLM serveCUDA metricsModelForge UIposture + galleryLLM gatewayenforce+recordApps selectFinOps meter

Glassbox thesis

Agents decide what to do. The model plane decides which weights, where, with what proof. This taxonomy replaces a mental map — every method and task type shows status, platform, and honest limits.

  • DomainForge — train (QLoRA/DPO) + S0→S4 eval ladder
  • ModelForge — posture, CUDA receipts, SLM bake-off
  • Classical ML — playbook MLOps stack (registry → CI/CD)
ModelForge taxonomy tab →DomainForge taxonomy tab →
Model Plane panel questions →Question bank (GitHub)

Status legend

  • Live demoLive demo
  • CUDA / artifact receiptCUDA / artifact receipt
  • Educational (not prod SLO)Educational (not prod SLO)
  • Playbook / interview archetypePlaybook / interview archetype
  • Planned — not claimed shippedPlanned — not claimed shipped

Eval regression in this org means golden-fixture CI gates — not sklearn linear regression. Tabular classification/regression live in the lane as interview-grounded design depth.

ADR-034 · ModelForge model plane · taxonomy UI plan