01
Week
Dev environment (uv, Ruff, Docker), git/PR workflow and review etiquette, FastAPI + Pydantic service, Postgres in Docker, first reviewed PR merged.
Curriculum
AI Backend and Full-Stack tracks. One 12-week roadmap.
The two tracks run as separate teams with separate codebases, integrating over a typed API contract, the same way real product orgs split frontend and AI/platform. Every 1 to 3 days a deliberately vague feature request lands; your squad writes the spec, branches, opens a PR, and the strongest PR gets merged into the live product. From week 5 the system takes real punishment: simulated production traffic and live security attacks. In week 12 it goes public with paid acquisition.
Each week repeats: RequestSpecBuildReviewMerge
12 weeks
Onboarding through public launch
2 tracks
AI Backend and Full-Stack
1 product
Shipped to real users in week 12
AI Backend Engineering
Production AI backend engineering
You graduate ready to design, ship, and operate the backend systems that power AI products in production.
Week by week
Onboarding
Week 1
Foundations
Weeks 2-4
02
Week
Self-hosting open-source / closed-source model alternatives with vLLM / Ollama (or the OpenAI API), OpenAI-compatible endpoints, quantization, streaming over SSE, structured outputs. First LLM-backed endpoint.
03
Week
Retrieval foundations: open-source / closed-source embedding model alternatives (BGE/E5/Nomic), chunking strategies, pgvector / Pinecone. Postgres deep-dive: HNSW vs IVFFlat indexes, query planning.
04
Week
Hybrid retrieval (BM25 + dense) with Qdrant / Pinecone, cross-encoder reranking, RAG evaluation with Ragas + Langfuse / LangSmith, Redis caching. A measured, working retrieval pipeline.
Acceleration
Weeks 5-8
05
Week
LangGraph / LangChain foundations: graphs, state, tool-use, structured outputs. Your first agent that actually does something.
06
Week
Multi-agent systems: supervisor/worker patterns, agent handoffs, parallelism, human-in-the-loop, Postgres-backed checkpointing and persistence.
07
Week
Durable execution with Temporal / managed workflow services (long-running workflows, retries, idempotency), Redis-backed jobs. First traffic simulation: load-test with k6 / Locust, hunt bottlenecks.
08
Week
Deep observability: Langfuse / LangSmith tracing, token/cost accounting, Prometheus + Grafana / Datadog infra metrics. First attacks: prompt injection and jailbreaks, then build input/output guardrails.
Hardening
Weeks 9-11
09
Week
Serving under real load: vLLM / OpenAI batching and concurrency, PgBouncer pooling, caching strategy, Redis rate limiting, latency/cost optimization.
10
Week
Security hardening: prompt-injection and RAG-poisoning defense, token-flooding/cost-abuse protection, PII handling, auth on endpoints. Defend against live attacks.
11
Week
Resilience: edge DDoS mitigation (with the infra/frontend team), graceful degradation, circuit breakers, fallbacks, Postgres backup/replication. Sustained load + attack drills, feature freeze.
Launch
Week 12
12
Week
Deploy on our own infrastructure with Docker Compose / k3s and Caddy + TLS, and set up CI/CD plus the supporting infrastructure as a team. Monitoring live, on-call rotation, real users from ad campaigns, incident response, retro.
Full-Stack Engineering
Modern full-stack product engineering
You graduate ready to ship production web applications with streaming, real-time, and AI-native user experiences.
Week by week
Onboarding
Week 1
01
Week
Bun, Next.js 15 App Router, TS strict, Biome, Tailwind v4, shadcn/ui. git/PR workflow. Postgres in Docker, Drizzle schema + first migration, first typed query shipped.
Foundations
Weeks 2-4
02
Week
RSC vs client components, Server Actions, data fetching and mutations, Suspense + streaming UI. Drizzle deep-dive: relations, migrations, query building.
03
Week
Forms and validation (Zod, useActionState, useOptimistic), loading/error states, TanStack Query client cache, auth with Better Auth / Clerk (sessions, providers, protected routes).
04
Week
Integrating the AI backend: typed contracts, streaming AI responses with the Vercel AI SDK / provider APIs, generative UI, optimistic updates. PgBouncer pooling + Redis caching.
Acceleration
Weeks 5-8
05
Week
Advanced rendering: Partial Prerendering, edge vs node runtime, Next caching and revalidation tags, image/font optimization, performance budgets.
06
Week
Real-time: WebSockets/SSE, live updates, presence, optimistic concurrency. State with Zustand/Jotai, polished interaction design with Motion.
07
Week
First traffic simulation: Web Vitals under load, k6 load tests, hunting render and DB bottlenecks, killing N+1 queries in Drizzle, caching wins.
08
Week
Observability: error tracking with GlitchTip / Sentry, structured logging, Web Vitals monitoring. First attacks: XSS, CSRF, auth abuse, then build defenses.
Hardening
Weeks 9-11
09
Week
Performance at scale: RSC streaming tuning, query optimization, Redis caching, static/CDN strategy on self-hosted infra, bundle analysis.
10
Week
Security hardening: input sanitization, CSP, rate limiting, bot/abuse defense, session and token hardening, secure headers. Defend live attacks.
11
Week
Resilience: edge DDoS mitigation (Nginx/Caddy, with the backend team), error boundaries, graceful degradation, accessibility pass, feature freeze.
Launch
Week 12
12
Week
Launch on our own infrastructure: standalone Next.js, Docker, Bun/Node behind Caddy + TLS, with CI/CD and supporting infrastructure set up by the team (not a managed platform like Vercel). Monitoring live, real users from ad campaigns, incident response, retro.