What you’ll ship

You don’t finish with a certificate. You finish with these.

Every course ends in a real, code-verified artifact — the kind of thing you put in front of a hiring manager and say “run it yourself.” Here's what stacks up in your portfolio.

decision-memo-07.md

# Duplicate charge — retry storm
## Decision: idempotency key on charge
- Option A: dedupe in the worker  ✗ races
- Option B: idempotency key       ✓ chosen
Invariant: one order → at most one charge

Foundations · Engineering Judgment

A decision memo

You can turn a 3am incident into a decision a reviewer can inspect — options, tradeoffs, the call, and why.

charge.test.ts

POST /charge  (retried 3×)
expect(charges).toHaveLength(1)
✓ 1 charge for 1 order
✓ safe to retry (idempotency-key)
PASS  12/12 checks

Engineering · Backend

An idempotent API

You can build a service that survives retries, partial failures, and real load — verified by a check, not a claim.

schema-review.sql

orders ─┐
        ├─< order_items >─ products
payments ┘  (FK, ON DELETE RESTRICT)
UNIQUE(order_id, idempotency_key)
✓ survives review · no orphan rows

Data · Databases

A schema under review

You can model data that stays correct at scale — and defend every constraint when a senior pushes back.

rag-eval.json

query → retrieve(k=5) → generate
eval: faithfulness  0.92  ✓
eval: answer-relevance 0.88 ✓
eval: no-hallucination  PASS
golden set: 24/24 above bar

AI Engineering · RAG & Eval

A RAG feature + eval harness

You can ship an AI feature that answers from real sources — and prove it works with a rubric and a golden set.

design-doc.md

client → LB → api ×N → queue → workers
            └→ cache (read-through)
bottleneck: write amplification on fanout
tradeoff: consistency ↔ latency (chose SWR)
✓ defends under load + failure

Architecture · System Design

A system design you can defend

You can draw the system, name the bottleneck, and defend the tradeoffs — the exact thing a design round tests.

postmortem-2029-03.md

Impact: 4m 12s p99 > 2s on /checkout
Root cause: unbounded N+1 on cart load
Fix: batch + index (verified in staging)
Prevent: perf budget in CI (fails > 200ms)
✓ class of bug closed, not just this one

Reliability · Observability

An incident postmortem

You can make a failure a location — root-cause it, prevent the class, and write it up so it never recurs.

Every artifact lands in one verifiable portfolio.

Each passing proof stacks into an evidence ledger with a public link — the certificate is a curl command anyone can run, not a PDF they take on faith.

Start building yours See the method →

What you’ll ship

You don’t finish with a certificate. You finish with these.

Every course ends in a real, code-verified artifact — the kind of thing you put in front of a hiring manager and say “run it yourself.” Here's what stacks up in your portfolio.

decision-memo-07.md

# Duplicate charge — retry storm
## Decision: idempotency key on charge
- Option A: dedupe in the worker  ✗ races
- Option B: idempotency key       ✓ chosen
Invariant: one order → at most one charge

Foundations · Engineering Judgment

A decision memo

You can turn a 3am incident into a decision a reviewer can inspect — options, tradeoffs, the call, and why.

charge.test.ts

POST /charge  (retried 3×)
expect(charges).toHaveLength(1)
✓ 1 charge for 1 order
✓ safe to retry (idempotency-key)
PASS  12/12 checks

Engineering · Backend

An idempotent API

You can build a service that survives retries, partial failures, and real load — verified by a check, not a claim.

schema-review.sql

orders ─┐
        ├─< order_items >─ products
payments ┘  (FK, ON DELETE RESTRICT)
UNIQUE(order_id, idempotency_key)
✓ survives review · no orphan rows

Data · Databases

A schema under review

You can model data that stays correct at scale — and defend every constraint when a senior pushes back.

rag-eval.json

query → retrieve(k=5) → generate
eval: faithfulness  0.92  ✓
eval: answer-relevance 0.88 ✓
eval: no-hallucination  PASS
golden set: 24/24 above bar

AI Engineering · RAG & Eval

A RAG feature + eval harness

You can ship an AI feature that answers from real sources — and prove it works with a rubric and a golden set.

design-doc.md

client → LB → api ×N → queue → workers
            └→ cache (read-through)
bottleneck: write amplification on fanout
tradeoff: consistency ↔ latency (chose SWR)
✓ defends under load + failure

Architecture · System Design

A system design you can defend

You can draw the system, name the bottleneck, and defend the tradeoffs — the exact thing a design round tests.

postmortem-2029-03.md

Impact: 4m 12s p99 > 2s on /checkout
Root cause: unbounded N+1 on cart load
Fix: batch + index (verified in staging)
Prevent: perf budget in CI (fails > 200ms)
✓ class of bug closed, not just this one

Reliability · Observability

An incident postmortem

You can make a failure a location — root-cause it, prevent the class, and write it up so it never recurs.

Every artifact lands in one verifiable portfolio.

Each passing proof stacks into an evidence ledger with a public link — the certificate is a curl command anyone can run, not a PDF they take on faith.

Start building yours See the method →

You don’t finish with a certificate. You finish with these.

One system. Keep exploring.

You don’t finish with a certificate. You finish with these.

One system. Keep exploring.