rowsandbox / curriculum

Sixteen weeks, row by row

Four arcs, sixteen ships, nine AI-era modules. Each week is a row you can audit — what you learned, what you shipped, what you broke.

02
table: curriculum · 16 rows

Sixteen weeks, row by row

Four arcs. Sixteen ships. Each week is a row you can audit — what you learned, what you shipped, what you broke.

WHERE arc = * → 16 rows
W·01
Foundations

Python, seriously

Type hints, async, context managers, data structures & algorithms, and the stdlib you skipped. Benchmark, profile, stop writing notebooks that mutate globals.
ShipTyped CLI streaming 5GB in <512MB
W·02
Foundations

SQL that scales

Window functions, CTEs, query plans, why your cost is quadratic. Postgres today, Synapse dedicated pools Friday.
ShipQuery rewrite cutting runtime 80%+
W·03
Foundations

Git, containers, CI/CD

Bicep and Terraform. Docker fundamentals. GitHub Actions pipelines that fail loudly. Secrets that don't leak.
ShipReproducible Azure env in one command
W·04
Foundations

Storage & the lakehouse

OLTP vs OLAP. Star and snowflake schemas. ADLS Gen2, Delta Lake, partitioning, Z-ordering. Warehouse vs lake vs lakehouse vs mesh.
ShipBronze/Silver/Gold + schema contracts
W·05
Pipelines

PySpark fluency

DataFrame APIs, UDFs (and when not), broadcast joins, skew, reading SparkUI like a radiologist — on Synapse and Databricks clusters alike.
Ship100M-row job in <20 min
W·06
Pipelines

Synapse & Databricks

Synapse notebooks, pipelines, managed identities — then the same workloads on Azure & AWS Databricks: Unity Catalog, Jobs, Delta Live Tables. Pick per workload, not per religion.
ShipSame pipeline on Synapse + Databricks
W·07
Pipelines

Orchestration & streaming

Airflow, ADF, Dagster compared. Messaging — Kafka, RabbitMQ, SQS — and when async beats batch. Alerts that wake you only when they should.
ShipDAG with SLA monitors + auto-recovery
W·08
Pipelines

Quality & governance

Great Expectations, dbt tests, Purview. Data contracts between producers and consumers. Lineage and GDPR posture that holds up in legal review.
ShipContract-validated pipeline · full lineage
W·09
AI Systems

Embeddings & vector stores

Semantic search, recommendations, anomaly detection. Chunking strategies. Azure AI Search, Bedrock Knowledge Bases, Pinecone, Qdrant, pgvector — when each fits.
ShipHybrid BM25+vector index over 50K docs
W·10
AI Systems

Retrieval that works

Semantic ranker, MMR, rerankers, metadata filters, and the failure modes of every naive RAG demo. Evaluate with RAGAS.
ShipRetrieval pipeline + reranker + evals
W·11
AI Systems

Prompt engineering & safety

Prompts as versioned, evaluated code: system design, structured outputs, tool calling, model routing across Azure OpenAI and AWS Bedrock. Injection defense, moderation, adversarial testing.
ShipStructured-output service <800ms p95
W·12
AI Systems

Multimodal & document AI

OCR with Azure Document Intelligence. Vision APIs for layouts. Whisper for speech. Multi-model agreement rules that survive audit.
Ship10K-PDF extraction pipeline + audit trail
W·13
AI Systems

Agentic AI & MCP

Azure AI Foundry agents, AWS Bedrock Agents, LangGraph state machines. Model Context Protocol tool servers. Planner/executor and multi-agent orchestration — and when a plain pipeline still wins.
ShipMulti-agent system · 3 MCP servers · HITL gate
W·14
Ship

Deployment & APIM

FastAPI backends. Static Web Apps. API Management as governance plane. Cosmos DB threads. The real architecture behind "AI at work".
ShipProduction AI service + rate limits
W·15
Ship

Capstone build week

Heads down. Pair reviews. Architecture critiques. We break your thing on purpose so the board can't.
ShipEnd-to-end capstone v1 + eval suite
W·16
Ship

Demo day & defense

Present to a panel of working engineers — not recruiters. Hard questions about observability, failure modes, cost per inference. You answer them.
ShipSigned, repo-complete capstone + runbook
+
table: ai_era_modules · 9 rows · woven through W09–W16

Built for the AI era

These aren’t electives bolted on at the end — they’re disciplines threaded through the second half of the program. Every module ships against real systems on Azure and AWS.

M·01

Prompt Engineering

System prompts as versioned code. Structured outputs, few-shot design, eval-driven iteration, model routing. You stop guessing and start measuring.

Azure OpenAI · Bedrock · promptfoo
M·02

Agentic AI

Planner/executor patterns, state machines, multi-agent orchestration, human-in-the-loop gates — and the honest cases where a pipeline beats an agent.

AI Foundry · Bedrock Agents · LangGraph
M·03

MCP & Tool Use

Model Context Protocol servers, function calling, tool registries, auth boundaries. Give models hands without giving them the keys.

MCP · function calling · OAuth scopes
M·04

RAG Systems

Chunking, hybrid retrieval, rerankers, citation discipline. The failure modes of every naive demo — and how to engineer past them.

AI Search · Bedrock KBs · pgvector
M·05

LLMOps & Evals

RAGAS, LLM-as-judge, regression suites, canary prompts, cost-per-inference budgets. If you can’t measure it, you can’t ship it.

RAGAS · OTel · App Insights
M·06

Multimodal AI

Vision extraction for layouts and signatures, Whisper for speech, image-generation guardrails. Document AI built to audit grade.

Doc Intelligence · GPT-4o Vision · Whisper
M·07

Fine-Tuning & Routing

LoRA adapters, distillation, model routers. When a small fine-tuned model beats a frontier call — in cost, latency, and privacy.

LoRA · Model Router · HF Hub
M·08

AI Safety & Governance

Prompt-injection defense, content moderation, red-teaming, audit trails. Posture mapped to the EU AI Act and NIST AI RMF — in language legal will accept.

Content Safety · Purview · NIST RMF
M·09

AI-Assisted Engineering

Pair-building with coding agents: spec-first prompting, review discipline, agentic workflows in your repo. Ship faster without shipping slop.

Claude Code · Copilot · spec-driven dev
05
table: capstones · 5 rows

Ship one. Or pitch your own

Four reference capstones with real scope and real stakes. Bring your own with instructor approval — we've shipped medical, legal, fintech, and logistics variants.

OP·01
Legal · Claims

A defensible claims-processing pipeline

10,000+ PDFs through Document Intelligence and GPT-4o Vision with 2-of-3 agreement on signature detection. Every decision carries a page reference, a confidence score, and a reproducible trace.
throughput 2.4K pages/min
audit 100% traceable
stack Synapse · ADI · AI Search
OP·02
Finance · Logs

AI log analytics for regulated banking

Vector-indexed operational logs with anomaly retrieval agents. Shipped on Azure AI Search or AWS S3 Vectors + Bedrock Agents — both reference architectures provided.
scale 10M events/day
retrieval hybrid + reranker
governance Purview · lineage-first
OP·03
Health · PubMed

Medical literature query agent

PubMed corpus chunked for clinical precision, surfaced through a tool-using agent that refuses to answer outside its citations.
corpus 35M abstracts
eval RAGAS + clinician review
OP·05
Ops · Agentic

Agentic ops copilot for a data platform

A multi-agent system that triages pipeline failures, queries lineage over MCP, drafts the fix as a pull request, and waits for a human to approve. LangGraph orchestration across Foundry and Bedrock models.
agents 4 · planner/executor
tools 3 MCP servers
guardrails HITL approvals + evals
OP·04
Internal · Concierge

An AI concierge for your company

FastAPI backend, Azure Static Web App front, APIM governance, Cosmos DB thread memory. The architecture actually serving enterprise users today — not a demo.
arch FastAPI · SWA · APIM · Cosmos
latency <800ms p95
obs App Insights + OTel
applications · C04 · open

Ready to transform your career?

Join thousands of professionals who have successfully transitioned into high-paying data careers. Cohort 04 start date: TBD — applications reviewed on a rolling basis.

Apply now — limited spots Schedule free consultation