Build These First¶
Ten projects ordered by difficulty. Each maps to specific modules so you know what to learn before you build.
Portfolio rule of three
Ship one RAG app, one agent, and one production-ready demo with evals — that's stronger than finishing every lesson without building.
Related: Course 16 · Capstones (full briefs) · Exercises (skill drills)
At a glance¶
| # | Project | Difficulty | Modules | Est. time |
|---|---|---|---|---|
| 1 | Doc Q&A bot (RAG starter) | Beginner | Course 02, Course 06 | 1 weekend |
| 2 | Enterprise RAG with citations | Intermediate | Course 06, Course 10, Course 12 | 1–2 weeks |
| 3 | Semantic search over code/docs | Intermediate | Course 06, Course 10 | 1 week |
| 4 | Tool-using research agent | Intermediate | Course 07, Course 08 | 1–2 weeks |
| 5 | Multi-agent research system | Advanced | Course 09, Course 06, Course 13 | 2–3 weeks |
| 6 | Support bot with routing | Intermediate | Course 06, Course 07, Course 09 | 2 weeks |
| 7 | LLM data extraction pipeline | Intermediate | Course 11, Course 12 | 1 week |
| 8 | Domain style fine-tune | Advanced | Course 15, Course 05 | 2–3 weeks |
| 9 | AI quality eval suite | Intermediate | Course 13, Course 14 | 1 week |
| 10 | Deploy your AI app | Advanced | Course 12, Course 16 | 1–2 weeks |
1. Doc Q&A bot (RAG starter)¶
What: Upload PDFs, ask questions, get answers grounded in your docs.
Learn first:
| Module | Lessons |
|---|---|
| Course 02 | L2 First app, L5 APIs |
| Course 06 | L1–5 (intro → basic RAG) |
Build checklist:
- [ ] Chunk documents (500–1000 tokens)
- [ ] Embed + store in Chroma or FAISS
- [ ] Retrieve top-k, inject into prompt
- [ ] Show source snippets in UI
Stretch: Add Streamlit or FastAPI frontend.
Capstone link: Capstone project 1 — RAG Knowledge Assistant
2. Enterprise RAG with citations¶
What: Production-style RAG with hybrid search, citation links, and basic monitoring.
Learn first:
| Module | Lessons |
|---|---|
| Course 06 | L6–10 (advanced RAG, hybrid, eval, production) |
| Course 10 | L1–5 (indexing, schema) |
| Course 12 | L2 Observability |
Build checklist:
- [ ] Hybrid BM25 + vector search
- [ ] Reranker or score threshold
- [ ] Inline citations
[doc_id:chunk] - [ ] Log retrieval scores + latency
Capstone link: Capstone project 1 (extended)
3. Semantic search engine¶
What: Search interface over a corpus (docs, blog, codebase) with filters and highlighting.
Learn first:
| Module | Lessons |
|---|---|
| Course 06 | L2 Vector DBs, L4 Retrieval |
| Course 10 | L6–8 (schema, scaling, hybrid) |
Build checklist:
- [ ] Metadata filters (date, tag, author)
- [ ] Highlight matched spans
- [ ] Evaluate recall@k on 20 hand-labeled queries
Capstone link: Capstone project 6 — Semantic Search Engine
4. Tool-using research agent¶
What: Agent that searches the web (or files), summarizes, and cites sources.
Learn first:
| Module | Lessons |
|---|---|
| Course 07 | L1–4, L7 (loop, ReAct, tools) |
| Course 08 | L1–3 (harness, loop, tools) |
Build checklist:
- [ ] ReAct loop with max 10 steps
- [ ] 2–3 tools (search, read_file, summarize)
- [ ] Structured final report (markdown)
- [ ] Trace log of each step
Capstone link: Capstone project 2 — Autonomous Coding Agent (adapt tools for research)
5. Multi-agent research system¶
What: Orchestrator delegates to specialist agents (searcher, analyst, writer).
Learn first:
| Module | Lessons |
|---|---|
| Course 09 | L1–6 (orchestrator, handoffs) |
| Course 06 | L9 Agentic RAG |
| Course 13 | L4 Agent trajectory evals |
Build checklist:
- [ ] Orchestrator + 2 worker agents
- [ ] Shared state or blackboard
- [ ] Eval: did final report cite all sources?
Capstone link: Capstone project 3 — Multi-Agent Research
6. Support bot with routing¶
What: Customer support bot that routes billing vs technical questions to different knowledge bases.
Learn first:
| Module | Lessons |
|---|---|
| Course 06 | L1–5 |
| Course 07 | L10 Workflow vs Agent |
| Course 09 | L5 Supervisor/Router |
Build checklist:
- [ ] Router classifies intent
- [ ] Separate RAG indexes per domain
- [ ] Escalation to human when confidence low
- [ ] Guardrails (Course 14 L4)
7. LLM data extraction pipeline¶
What: Extract structured JSON from invoices, emails, or PDFs at scale.
Learn first:
| Module | Lessons |
|---|---|
| Course 11 | L4 Structured output, L6 Chaining |
| Course 12 | L4 Caching, L8 API design |
Build checklist:
- [ ] JSON schema validation
- [ ] Batch processing with retries
- [ ] Golden set of 50 examples + accuracy metric
Capstone link: Capstone project 7 — Data Extraction Pipeline
8. Domain style fine-tune¶
What: Fine-tune (LoRA) a small model to match your team's writing style or report format.
Learn first:
| Module | Lessons |
|---|---|
| Course 05 | L6–7 Fine-tuning, instruction tuning |
| Course 15 | L1–5 (when, data prep, LoRA) |
Build checklist:
- [ ] 200+ high-quality (input, output) pairs
- [ ] LoRA fine-tune on 7B or smaller
- [ ] Compare base vs fine-tuned on held-out set
- [ ] Document when RAG would have been enough (FAQ)
9. AI quality eval suite¶
What: Automated eval pipeline for an LLM app — golden dataset, LLM-as-judge, CI gate.
Learn first:
| Module | Lessons |
|---|---|
| Course 13 | All 6 lessons |
| Course 14 | L3 Hallucination, L8 Red teaming |
Build checklist:
- [ ] 30+ golden Q&A pairs
- [ ] Regression test in CI
- [ ] Dashboard or report for pass/fail trends
Capstone link: Capstone project 9 — AI Safety Evaluation Suite
10. Deploy your AI app¶
What: Take project #1, #2, or #4 to production — Docker, env config, monitoring, cost caps.
Learn first:
| Module | Lessons |
|---|---|
| Course 12 | L6–10 (cost, deployment, scaling) |
| Course 13 | L5–6 (CI/CD, monitoring) |
Build checklist:
- [ ] Dockerfile + health check
- [ ] Secrets via env vars
- [ ] Rate limiting + cost alerts
- [ ] README with architecture diagram
Capstone link: Capstone project 10 — Deploy
Suggested build order by persona¶
| Persona | Build order |
|---|---|
| Complete beginner | 1 → 4 → 9 |
| Software engineer | 1 → 2 → 4 → 10 |
| ML engineer | 2 → 5 → 8 |
| Career switcher | 1 → 4 → 6 → 9 → 10 |
Start routing: Start Here