Os Labs · uma oficina prática

Uma oficina, não uma planilha.

Sistemas reais que você constrói e mantém — cada um com uma especificação, uma prova executável e a linha exata que você pode colocar no seu currículo. Escolha um projeto, entregue-o e saia com evidências.

projetos, e contando

trilhas

137

verificações de prova executáveis

Mensal

novos projetos adicionados

Como um projeto funciona

Todo lab termina em algo que um cético pode re-executar.

01 · frame

A real brief

Every lab opens with a spec a real team would recognize — not a puzzle with one clever answer.

02 · build

You build it

A starter scaffold and the Sage loop: map the system, decide under tradeoffs, ship the thing.

03 · prove

It passes checks

Each build ends in runnable acceptance checks — the proof a skeptic can re-run, the reason it belongs on your résumé.

atéSenior

AI EngineeringAdvanced~6h

Add a cross-encoder re-ranker to a RAG pipeline and prove it improves recall@k

Retrieve wide with embeddings, then re-rank with a real cross-encoder and measure the lift, not just eyeball it.

Linha do currículo

Built a two-stage RAG retrieval pipeline (bi-encoder recall + cross-encoder re-ranking) and measured a recall@5 and MRR improvement against a labeled eval set of 40+ query-passage pairs.

frame → route → prove4 verificações · prova

AI EngineeringWorking~5h

Build a semantic cache for LLM calls and prove the cost, latency, and correctness tradeoffs

Cache by meaning, not exact string match — then prove it actually saves money without serving stale garbage.

Linha do currículo

Built a semantic LLM response cache with embedding-similarity keying and TTL/manual invalidation, measuring a 40%+ cache hit rate and cost/latency reduction across a 100-query mixed workload while catching over-eager false-positive cache hits.

frame → route → decide → prove4 verificações · prova

AI EngineeringAdvanced~7h

Build a hallucination detector that flags ungrounded claims in LLM answers

Split the answer into claims, check each one against the retrieved context, and catch what the model made up.

Linha do currículo

Built a claim-level hallucination detector for a RAG pipeline that decomposes answers into atomic claims and checks each against retrieved context via entailment, catching 85%+ of injected false claims across a 30-example adversarial test set with a measured false-positive rate on true claims.

frame → route → decide → prove4 verificações · prova

AI EngineeringWorking~5h

Build a function-calling router that rejects bad arguments before any tool runs

Route the request to the right tool, validate every argument against its schema, and never let a malformed call through.

Linha do currículo

Built a function-calling router with JSON-schema argument validation across 5+ tools, blocking 100% of malformed tool calls pre-execution and driving a self-correction retry loop, verified against a 25-case test suite of ambiguous and adversarial routing requests.

frame → route → decide → prove4 verificações · prova

BackendWorking~8h

Signature-Verified Webhook Receiver

Accept third-party webhooks without trusting the network.

Linha do currículo

Built a webhook receiver with HMAC signature verification, timestamp-based replay protection, and idempotent event processing, backed by automated tests for forged and replayed payloads.

route → decide → prove4 verificações · prova

BackendWorking~7h

Cursor-Based Pagination Under Live Writes

Paginate a list that keeps changing underneath you.

Linha do currículo

Implemented cursor-based pagination with opaque tie-breaker cursors, replacing offset pagination to eliminate duplicate and skipped rows under concurrent inserts, verified with a concurrent-write test harness.

route → map → prove4 verificações · prova

DataWorking~7h

Cache-Aside Read Path with Correct Invalidation

Speed up reads without ever serving stale data.

Linha do currículo

Built a cache-aside layer with write-path invalidation and TTL fallback, eliminating stale reads under concurrent read/write load, validated with a race-condition test suite against Redis and PostgreSQL.

map → decide → prove4 verificações · prova

DataWorking~8h

Data Quality Gate for Batch Ingestion

Stop bad data before it reaches the warehouse.

Linha do currículo

Built a schema-and-constraint data quality gate for batch ingestion (type, null, range, and uniqueness checks) that blocks corrupt batches from loading and emits a row-level failure report, cutting bad-data incidents at the load boundary.

decide → prove4 verificações · prova

FrontendAdvanced~10h

Virtualized List Rendering 10k+ Rows at 60fps

Windowed rendering that stays smooth at 10,000+ rows.

Linha do currículo

Built a from-scratch virtualized list renderer (no react-window) handling 10k+ rows with sub-16ms frame times, verified with Chrome DevTools performance traces and a Playwright scroll-jank regression test.

frame → decide → prove4 verificações · prova

FrontendWorking~6h

Accessible Modal with a Real Focus Trap

A dialog that keyboard and screen-reader users can actually use.

Linha do currículo

Implemented a WCAG 2.2-compliant modal dialog (ARIA dialog pattern, focus trap, focus restoration) with automated axe-core scans and NVDA/VoiceOver manual verification scripts, eliminating a common accessibility failure class.

frame → decide → prove4 verificações · prova

QualityAdvanced~9h

Consumer-Driven Contract Test Suite

Contracts that fail the build the moment an API breaks a consumer.

Linha do currículo

Set up consumer-driven contract testing (Pact) between frontend and API teams, catching a deliberately introduced breaking response-shape change in CI before deployment — replacing brittle end-to-end integration tests with fast, isolated contract verification.

frame → map → decide → prove4 verificações · prova

QualityAdvanced~8h

p99 Load Test with a Latency Budget Gate

A load test that fails the build when p99 latency blows the budget.

Linha do currículo

Built an automated p99 latency budget gate (k6 + CI) that load-tests critical API endpoints under realistic concurrency and fails the pipeline on SLO violations, catching a deliberately introduced N+1 query regression before merge.

frame → decide → prove4 verificações · prova

CloudAdvanced~6h

Blue-Green Deployment with Health-Gated Cutover

Deploy a green environment, gate traffic on real health checks, and roll back in one command.

Linha do currículo

Built a zero-downtime blue-green deployment pipeline with automated health-gated traffic cutover and sub-5-second rollback, verified with a chaos test that killed a bad deploy before it received a single request.

frame → route → decide → prove4 verificações · prova

SecurityWorking~5h

Role-Based Access Control, Enforced at the Boundary

Define roles and permissions once, enforce them at every route, and prove a lower-privilege user gets denied.

Linha do currículo

Designed and implemented a role-based access control layer with centralized middleware enforcement across 8+ protected endpoints, closing a privilege-escalation gap where lower-role users could previously call admin-only mutations directly.

frame → decide → prove4 verificações · prova

SecurityWorking~4h

An Input Sanitizer That Actually Stops XSS

Escape and allowlist untrusted HTML at the boundary, then prove a real stored-XSS payload renders inert.

Linha do currículo

Closed a stored-XSS vulnerability in a user-generated-content field by replacing ad-hoc regex escaping with an allowlist-based sanitizer, verified against a 12-payload OWASP-derived attack corpus with zero script execution.

frame → decide → prove4 verificações · prova

FoundationsWorking~4h

Turn Recursion Into Iteration Without Losing Correctness

Convert a stack-overflowing recursive function into an iterative one with an explicit stack, and prove it survives inputs the original crashed on.

Linha do currículo

Rewrote a recursive tree-traversal algorithm to use an explicit stack after profiling a production stack-overflow on deeply nested input, eliminating the crash while proving output-identical behavior against the original across 500+ generated test cases.

frame → decide → prove4 verificações · prova

AI EngineeringWorking~6h

Ground a chatbot in your own docs

A small RAG service: you index a folder of documents into a vector store, retrieve the top-k passages for a question, and have the model answer strictly from that context — citing the source and abstaining when the answer isn’t present. You finish with a service that turns "the model guessed" into "the model looked it up."

Linha do currículo

Built a retrieval-augmented Q&A service (chunking, embeddings, top-k retrieval, grounded generation with citations and abstention) and proved grounding with an automated faithfulness check.

frame → map → decide → prove3 verificações · prova

AI EngineeringWorking~5h

Build an eval harness for your AI

A test suite for an AI feature: a golden set of real inputs with known-good answers, a scorer (rubric or judge) that grades each output, and a single aggregate score. You change a prompt and watch the number move — the way a real team ships model changes without breaking things.

Linha do currículo

Built an evaluation harness (golden set, rubric/LLM-judge scoring, aggregate metric with regression threshold) that gates prompt/model changes on measured quality.

frame → decide → prove3 verificações · prova

BackendApplied~3h

A payment endpoint that can’t double-charge

A charge endpoint that stays correct when the network retries. You add an idempotency key, a store of handled keys, and a guard that makes a repeated request a no-op returning the original result — so three retries still charge the card once.