AI Engineering · LLM Application Developer — Learning Repository

🎓 Part of the free, open-source AI Career Curriculum ecosystem — Infrastructure · ML Engineering · AI Engineering · Governance. Live cohorts & team programs: ai-infra-curriculum.github.io.

💜 Sponsor this curriculum — sponsorships keep the whole open-source AI Career Curriculum free and moving.

Role level: 20 (application-developer craft — AI Engineering family)

Ship LLM-backed features end-to-end as an application developer: call hosted LLM APIs, engineer prompts and force schema-constrained structured output, declare and consume tools, orchestrate async / streaming calls with honest tail-latency reporting, keep cost defensible with prompt caching and cascade routing, ground answers with a minimal retrieval path, gate a feature on a golden-set regression suite, and land the whole thing behind a small HTTP API with feature flags, safe secrets, basic traces, a runbook, and input / output validation.

Status: mod-001 is shipped. The rest of the plan (mod-002 through mod-007 + two projects) is authored from .aicg/curriculum-plan.json and lands as autonomous content cycles run. See CURRICULUM.md for the full plan and JOB_REQUIREMENTS.md for the 31-posting sample that grounds the coverage decisions.

What you get

Planned commitment: 70 hours across 7 modules + 50 hours across 2 projects = ~120 hours.

• mod-001 — Prompt Engineering Foundations (shipped) — chat-style LLM APIs, prompt anatomy, structured JSON output with schema validation, token estimation, and the three most common shapes of prompt failure. Nine short exercises, one lab (text-classifier), and a quiz.
• mod-002 — Tool and Function Calling (planned) — declare typed tool schemas, run the tool_call → tool_result loop end-to-end, handle parallel tool calls and malformed arguments. Closes the "partial" coverage flagged in JOB_REQUIREMENTS.md (58% of postings).
• mod-003 — Streaming, Async, and Parallel LLM Orchestration (planned) — SSE stream consumption, partial JSON streaming, async fan-out with bounded concurrency, retry with backoff + jitter, honest p50 / p95 / p99 latency reporting.
• mod-004 — Model Selection, Cost, and Prompt Caching (planned) — cost-per-call estimation, small-vs-frontier A/B benchmarks, Anthropic and OpenAI prompt-caching with hit-ratio measurement, cascade routing, and graceful degradation for a provider outage.
• mod-005 — Retrieval Basics for LLM Applications (planned) — a minimal retrieval intro: embeddings, pgvector, top-k retrieval, grounded prompts with citations. Full RAG depth is delegated to rag-engineer-learning (level 30).
• mod-006 — Minimal Evaluation and Regression Checks (planned) — a minimal golden-set intro: 20 to 50 example fixtures, pytest CI gate, rule / similarity / LLM-as-a-judge scoring. Full evaluation engineering is delegated to ai-eval-engineer-learning (level 30).
• mod-007 — Shipping a First Production LLM Feature (planned) — FastAPI endpoint with streaming, safe secrets, safe traces, feature flag for provider / model swap, runbook, and input / output validation. Full production hardening (cloud deployment, CI/CD, production guardrails) is delegated to applied-ai-engineer-learning (level 30).
• project-001 — Multi-Provider Comparison Report (planned, 20h) — one feature, two providers, honest cost / latency / quality comparison.
• project-002 — Grounded, Tool-Calling, Streaming LLM Feature (planned, 30h) — production-shaped LLM feature end-to-end with SSE, tools, retrieval, caching, golden-set CI, traces, feature flag, and runbook.

Level ladder — where this track fits

• level 10 — ai-infra-junior-engineer-learning (assumed engineering-craft prerequisites)
• level 20 (peers) — llm-application-developer-learning (this track) and agentic-ai-developer-learning. The two split the level-20 LLM-application space: this track owns non-agentic patterns; the peer owns agent frameworks and multi-step agent orchestration.
• level 25 — ai-risk-engineer-learning (AI-risk-engineering craft — harm modelling, red-team, guardrail effectiveness)
• level 30 (next-up in this family) — rag-engineer-learning (full RAG pipeline), ai-eval-engineer-learning (application-layer evaluation + observability), applied-ai-engineer-learning (production hardening + cloud deployment), agentic-ai-engineer-learning (production agent engineering)
• level 40+ — senior-agentic-ai-engineer-learning and higher AI Engineering / Governance / Infra tracks

See CURRICULUM.md for the ownership rule this track applies and PREREQUISITES.md for the assumed lower-level curriculum.

Layout

llm-application-developer-learning/
├── lessons/mod-XXX-*/        modules with lectures, exercises, labs, quizzes
├── projects/project-XXX-*/   multi-module capstones
├── CURRICULUM.md             role-level coverage map
├── PREREQUISITES.md          assumed entry skills
├── VERSIONS.md               release history
├── JOB_REQUIREMENTS.md       requirements catalog with cited postings evidence
├── .aicg/job-requirements.json    machine-readable requirements & posting sample
├── .aicg/curriculum-plan.json     machine-readable module / exercise / project plan
└── README.md                 this file

Paired Solutions Repo

llm-application-developer-solutions carries the reference implementations.

---

Maintained by VeriSwarm.ai