Coverage v2 — Adding MLA-C01
Extends the Claude coverage to a three-cert program by folding in AWS MLA-C01 (ML Associate).
HungerSync Case Series — Adding MLA-C01 (Coverage v2)
Section titled “HungerSync Case Series — Adding MLA-C01 (Coverage v2)”Extends coverage-matrix.md and case-map.md to a three-certification program.
Verdict: not a pivot — additive. MLA-C01 adopts the one HungerSync subsystem the
GenAI certs structurally exclude (the predictive ML / prediction engine), giving the
world a clean three-layer stratification.
1. Why it fits (the exclusion proof)
Section titled “1. Why it fits (the exclusion proof)”- Claude Architect excludes training, fine-tuning, embeddings.
- AIP-C01 out-of-scope explicitly names “Model development and training” and “Data engineering and feature engineering.”
- MLA-C01 is exactly those: data prep, feature engineering, training, tuning, evaluation, deployment, drift, retraining.
The HungerSync prediction engine (delay / demand / taste / pre-positioning models) was always in the business architecture but had no cert home. MLA-C01 is that home.
2. Three layers, three lenses, one world
Section titled “2. Three layers, three lenses, one world”| Layer | Cert | HungerSync surface |
|---|---|---|
| Prediction (upstream ML/MLOps) | MLA-C01 | delay/demand/taste models: data→features→train→tune→eval→deploy→drift→retrain |
| Application (GenAI on AWS) | AIP-C01 | RAG, FM integration, safety, ops, deployment of the GenAI app |
| Agent (reasoning/orchestration) | Claude Architect | agentic loops, MCP, Claude Code, multi-agent dispatch |
The prediction layer feeds the agent layer. Upstream → downstream.
3. MLA-C01 at a glance (pass 720/1000; 50 scored Qs; MC + multi-response + ordering + matching)
Section titled “3. MLA-C01 at a glance (pass 720/1000; 50 scored Qs; MC + multi-response + ordering + matching)”| Domain | Weight |
|---|---|
| D1 Data Preparation for ML | 28% |
| D2 ML Model Development | 26% |
| D3 Deployment & Orchestration of ML Workflows | 22% |
| D4 ML Solution Monitoring, Maintenance & Security | 24% |
4. Net-new vs shared
Section titled “4. Net-new vs shared”Net-new (needs the predictive-ML lifecycle — these are the distinct MLA surface): D1 entirely (1.1 ingest/store, 1.2 transform/feature-eng, 1.3 integrity/bias) · D2 entirely (2.1 modeling approach, 2.2 train/tune/fine-tune, 2.3 evaluate) · D3.1 (ML endpoints, batch inference) · D3.3 retraining loop · D4.1 model drift.
Shared (absorb via tags on existing cases — do NOT duplicate): D3.2 (IaC/containers/autoscale) → CS1, CS6 · D3.3 generic CI/CD → CS7 · D4.2 (infra/cost monitoring) → CS9 · D4.3 (IAM/VPC/security) → CS8 · D1.3 PII/residency → CS8.
5. Three new predictive-ML cases
Section titled “5. Three new predictive-ML cases”| Case | Title | Layer | MLA tasks | Notes |
|---|---|---|---|---|
| ML1 | Will this flight be late? | Prediction | 1.1, 1.2, 1.3, 2.1 | Data foundation + feature engineering + model choice. Home for PiAware/weather edge ingestion (Kinesis/Kafka/Flink) + Feature Store. Resolves edge watch-item W4. |
| ML2 | Training, tuning, and trusting the model | Prediction | 2.2, 2.3 | Hyperparameter tuning (AMT), ensembling, fine-tuning the taste classifier (resolves W1), F1/AUC/RMSE, Clarify bias, shadow vs prod. |
| ML3 | Shipping and watching the model | Prediction | 3.1, 3.3, 4.1 | SageMaker endpoints + batch inference for the nightly forecast (ties to CS9) + SageMaker Pipelines retraining loop + Model Monitor drift. |
Three net-new cases cover an entire 4-domain cert because the rest is shared. That is as tight as a new layer can be.
6. Integrated lineup (12 cases + capstone) with tri-cert tags
Section titled “6. Integrated lineup (12 cases + capstone) with tri-cert tags”C = Claude Architect · A = AIP-C01 · M = MLA-C01
| # | Title | Primary | Also tagged |
|---|---|---|---|
| ML1 | Will this flight be late? | M | — |
| ML2 | Training, tuning, trusting the model | M | — |
| ML3 | Shipping & watching the model | M | A (batch/deploy overlap) |
| CS1 | Designing HungerSync | A | M (IaC), C (arch judgment) |
| CS2 | The ordering agent you can trust | C | A |
| CS3 | The storm breaks the pipeline | C+A | — |
| CS4 | Discovery you can trust | A | C |
| CS5 | The four-hour delay conversation | C | A |
| CS6 | Claude Code: config & workflows | C | M (IaC) |
| CS7 | Claude in the pipeline: review & CI | C | A, M (CI/CD) |
| CS8 | Trust & Safety: guardrails, privacy, governance | A | M (IAM/VPC/PII) |
| CS9 | The ops center | A+C | M (infra/cost/drift) |
| CS10 | Capstone: end-to-end ground stop | all three | spans every layer |
Layer balance: Prediction (M) = ML1–ML3; Application (A) = CS1/CS4/CS8/CS9; Agent (C) = CS2/CS5/CS6/CS7; spine joints = CS3/CS9; capstone = CS10.
7. MLA-C01 task → case coverage (all 12)
Section titled “7. MLA-C01 task → case coverage (all 12)”| MLA task | Covered by |
|---|---|
| 1.1 Ingest & store data | ML1 |
| 1.2 Transform & feature engineering | ML1 |
| 1.3 Data integrity, bias, PII/residency | ML1 (bias/integrity), CS8 (PII/residency) |
| 2.1 Choose modeling approach | ML1 |
| 2.2 Train & refine (tune, ensemble, fine-tune) | ML2 |
| 2.3 Analyze model performance | ML2 |
| 3.1 Select deployment infra (endpoints, batch) | ML3, CS9 |
| 3.2 Create/script infra (IaC, containers, autoscale) | CS1, CS6 |
| 3.3 CI/CD orchestration + retraining | ML3 (retraining), CS7 (CI/CD) |
| 4.1 Monitor model inference (drift) | ML3 |
| 4.2 Monitor & optimize infra & cost | CS9 |
| 4.3 Secure AWS resources (IAM/VPC) | CS8 |
All 12 covered. (Claude 30/30 and AIP 20/20 remain covered — unchanged.)
8. Watch items now resolved
Section titled “8. Watch items now resolved”- W1 (fine-tuning had no clean home) → ML2 (MLA 2.2 fine-tune pre-trained models).
- W4 (edge/IoT had no exam home) → ML1 (MLA 1.1 streaming ingestion + Feature Store).
Remaining: CS10 capstone confirmation; novel ordering (later).
9. Open decisions
Section titled “9. Open decisions”- Tighter still? ML1+ML2 could merge into one “build the delay model” case if you want 11 total instead of 12. (Lean: keep separate — data/modeling vs training/eval are distinct skill clusters and MLA weights D1+D2 at 54%.)
- Confirm CS10 as the tri-layer capstone.
- Build order unchanged: template → CS1 (reference) → outward; ML1 is the natural first case of the prediction track.