Playbook / Model Plane / Model Plane panel question bank (industry archetypes)

Model Plane panel question bank (industry archetypes)

Use with: Model Plane taxonomy · ModelForge · DomainForge · panel scripts

Sourcing honesty: Public prep guides (123ofAI, Interview Coder, ShadeCoder, SharpSkill, CalibreOS, vLLM docs/blog) converge on these archetypes for 2025–2026 Staff+ / LLM engineer loops. Few companies publish verbatim question lists for Principal model-plane roles — treat this as high-frequency practice priority, not a leaked question dump.

Evidence classes: O = org open proof · R = researched industry pattern · H = hypothetical numbers unless you defend a baseline · P = employer production (do not claim without primary records)


A. Model plane & strategy (CAIO / Principal)

#Question (as asked)What they probeStrong answer sketchProof
A1"You're agents-only — where's your model plane?"Orchestration vs weights/serve/economicsAgents call tools; model plane decides buy/RAG/PEFT/self-host with receipts — ModelForge posture + ADR-034O: ModelForge posture
A2"When do you fine-tune vs RAG vs prompt?"Decision tree before GPU spendPrompt first; RAG for changing facts + cites; PEFT for stable schema/behavior; buy when frontier reasoning winsO: ADR-019 DomainForge split
A3"Why not stuff SOPs into fine-tune weights?"Fact/behavior separationStale policies + hallucinated chunk_ids; RAG holds facts, PEFT holds JSON disciplineO: DomainForge S0→S4
A4"Show me you've run GPU training — not a Colab screenshot."CUDA receipt integritypeft_gpu.json with cuda=true; say T4 micro honesty; DomainForge 7B is depth pathO: peft_gpu.json
A5"When is self-host cheaper than API?"Cost/latency/residency mathSLM bake-off for schema tasks; vLLM receipt for tok/s; gateway FinOps — label H unless measuredO + H
A6"Design model routing for 50 tenant adapters."Multi-tenant serve economicsOne base + multi-LoRA vs N replicas; max_loras, eviction, sticky routingR: vLLM multi-LoRA docs
A7"How do apps pick models without N routers?"Federated planeApps select; gateway enforces + records (ADR-028/029)O: aegis-llm-gateway

Deep entry: 01 ModelForge flagship


B. LoRA · QLoRA · DPO · full fine-tune

#QuestionWhat they probeStrong answer sketchProof
B1"Explain LoRA — why is it cheap?"Math + intuitionFreeze W₀; train low-rank ΔW=BA; rank r ≪ d; ~100–1000× fewer trainable paramsR
B2"LoRA vs QLoRA?"Memory vs qualityQLoRA: 4-bit NF4 base + fp16/bf16 adapters; fits 65–70B on single 48–80GB GPU; slight train overheadR
B3"What are rank and alpha?"Hyperparameter literacyr = expressiveness; alpha scales adapter contribution; diminishing returns past r≈16–64 for many tasksR
B4"When does LoRA lose to full fine-tune?"Honest limitsHeavy domain shift / tasks needing large weight updates; often within 1–2 pts on standard tasksR
B5"Why DPO after SFT?"Alignment stageSFT teaches schema; DPO fixes plausible-wrong outputs via preference pairsO: DomainForge S4
B6"Can fine-tuning inject new facts?"RAG vs PEFT trapNo — use RAG; PEFT for format/intent/action codes, not memorizing policiesO: ADR-019
B7"Merge adapter vs serve dynamically?"Deploy trade-offMerge = zero adapter overhead; multi-LoRA = fleet economics, eviction riskR + O: Path B edu

Deep entry: 02 PEFT adaptation panel grill


C. Multi-LoRA · vLLM · inference serving

#QuestionWhat they probeStrong answer sketchProof
C1"Explain PagedAttention / KV cache."Inference fundamentalsFixed-size blocks; dynamic alloc; cuts fragmentation; enables prefix cacheR: vLLM paper/blog
C2"Continuous batching vs static?"ThroughputIteration-level scheduling; new requests join as others finishR
C3"How does multi-LoRA serving work in vLLM?"Fleet economicsShared base weights + per-request adapter; --enable-lora, --max-lorasR: vLLM LoRA docs
C4"Failure mode when max_loras exceeded?"Production scarAdapter eviction/reload latency; fix: working-set sizing, sticky routing, shard adaptersR
C5"TTFT vs throughput — what do you optimize?"SLO literacyInteractive: TTFT; batch/offline: throughput; cite receipt metrics honestlyO: vllm_cuda.json
C6"Path A CUDA vs Path B educational?"Honesty under grillPath B = mock adapter swap; Path A = upstream vLLM CUDA metrics — don't conflateO: ADR-022

Deep entry: 03 Multi-LoRA serving panel · ai-system-design/01 inference serving


D. Task types — classification · regression · JSON · eval

#QuestionWhat they probeStrong answer sketchProof
D1"Is your triage model 'classification'?"Task taxonomy precisionLLM JSON intent classification (~27 intents) — not sklearn tabular softmaxO: DomainForge
D2"Design tabular fraud classification at scale."Classical ML laneFeature store + registry + batch/online parity + champion/challengerR: mlops-llmops/04
D3"Regression vs classification in MLOps?"Classical vs LLM vocabTabular regression = numeric target + skew monitoring; eval regression = fixture CI gatesO: golden-eval
D4"How do you score format vs faithfulness?"Separate eval dimensionsschema_pass vs citation faithfulness — don't lump into one accuracyO: DomainForge eval
D5"When does an SLM beat GPT-4?"Bake-off disciplineStructured/schema tasks; cost/latency/privacy; public SLM tableO: ModelForge bake-off

Deep entry: 04 Task types & classical ML crossover


E. Classical MLOps crossover (Staff AI Platform)

#QuestionWhat they probeStrong answer sketchProof
E1"Design a model registry with promotion gates."Registry authorityImmutable digests; stage/alias; eval evidence required; rollback = retarget aliasR: mlops-llmops/01
E2"What is training–serving skew?"Feature paritySame feature name, different computation offline vs online; feature store fixesR: mlops-llmops/02
E3"Data drift vs concept drift?"Monitoring literacyInput distribution shift vs P(yx) change; PSI common for data drift
E4"When do you retrain after drift?"Gated promoteShadow/challenger beat champion + no contract regression + approver for risk classR
E5"How does LLM plane connect to classical MLOps?"Stack integrationAdapters in registry; gateway metering; golden-eval as CI gate bridging bothO

Deep entries: mlops-llmops/01–05


F. Trap questions (say these cleanly)

TrapWrong answerRight answer
"Have you trained your own LLM?""Yes, we fine-tune 7B from scratch""I adapt open bases with PEFT; I don't pretrain foundations — here's adapter receipt + eval Δ"
"Is Path B multi-LoRA production?""Yes, we serve adapters on GPU""Educational OpenAI-shaped swap — CUDA Path A is the metrics receipt"
"Put policies in QLoRA weights""Fine-tune on SOP PDFs""RAG for facts; PEFT for JSON discipline only"
"Eval regression = linear regression?"Confuse terms"Fixture/metric regression in CI — different from tabular regression models"
"Higher LoRA rank always better?""Use r=256 everywhere""Diminishing returns + VRAM; task-dependent; defend with ablation"

Rehearsal order (45 min panel prep)

  1. Open venkat-ai.com/model-plane — walk Adaptation + Task tabs (5 min)
  2. Cold-open ModelForge posture — say ready vs smoke (2 min)
  3. Drill A1, B2, C3, D1, E2 aloud (15 min)
  4. Close with buy/RAG/PEFT/self-host tree + three links (3 min)
  5. Optional: Practice Arena rubric for linked playbook entries