Foundations & measurement
Follow the question-to-answer pipeline, inspect real code, construct ground truth, and establish an MVP baseline.
AI Engineering Lab · 16-lesson course
Learn to build a reliable RAG chatbot in 30 days.
RAG Formation is a completed 30-day project-based course. It guides learners from a first question through retrieval, citations, refusal behavior, evaluation, security, latency, cost, and pilot readiness.
$ course --status
▸ format project-based learning
▸ length 30 days
▸ lessons16 short engineering lessons
▸ status complete
● build → measure → learnCourse roadmap
A 30-day path from the first retrieval test to a pilot checklist.
Each phase produces an artifact and uses its results to guide the next engineering decision.
Retrieval engineering
Compare semantic, keyword, and hybrid retrieval, then work through chunking, metadata, filters, query handling, and reranking.
Grounded answer generation
Design context budgets, evidence-only prompts, stable citations, refusal behavior, and hallucination controls.
Reliability & product readiness
Plan graceful degradation, latency, cost, observability, security, deployment boundaries, and pilot readiness.
Course syllabus
Build a reliable company RAG QA chatbot, one measured decision at a time.
RAG Formation is a self-paced, project-based course. Across 16 short lessons, you work from a sanitized company corpus and turn each investigation into a concrete engineering artifact.
Course mission
Build a technical prototype for employees, QA and regulatory teams, customer support, suppliers, and auditors. It should answer SOP and policy questions with citations, search supplier questionnaires, and refuse questions unsupported by the indexed documents. The chatbot is the worked example; the course teaches beginners how to build enterprise-grade RAG systems.
How the course works
- Learn only the concepts needed for the current task.
- Practice with the sanitized Meridian Labs demo corpus.
- Inspect evidence and evaluation results.
- Apply the insight to the next engineering decision.
Recommended order
The four phases build on each other, so 1 → 16 is the easiest first pass. The course is self-paced: jump to a lesson whenever you need to look something up, continue mid-course, or explore a topic early.
Phase 1 · Lessons 1–4
Foundations and measurement
Trace the question-to-answer pipeline, build ground truth, and establish an MVP baseline.
Phase 2 · Lessons 5–8
Retrieval engineering
Compare retrieval signals, then work through chunking, metadata, filters, query handling, and reranking.
Phase 3 · Lessons 9–12
Grounded answer generation
Design evidence budgets, evidence-only prompts, stable citations, refusal behavior, and hallucination controls.
Phase 4 · Lessons 13–16
Reliability and product readiness
Plan graceful degradation, latency, cost, observability, security, deployment boundaries, and pilot readiness.
Completion criteria
- Explain the complete RAG request path and diagnose retrieval versus generation failures.
- Run and trust the evaluation harness, reporting quality and latency metrics separately.
- Explain how chunking, metadata, hybrid search, context assembly, and prompting affect quality.
- Define citation and refusal behavior, then produce a prioritized MVP backlog from measured failures.
16 lesson library
16 lesson library
Sixteen short, project-based lessons with a concrete engineering artifact at every step.
01 · Foundations & measurement
From employee question to cited answer
A first RAG quality model and a quality cheat sheet.
Open lesson02 · Foundations & measurement
Trace one question through the real code
A one-page system trace for one question.
Open lesson03 · Foundations & measurement
Build a trustworthy evaluation set
An evaluation-set design checklist.
Open lesson04 · Foundations & measurement
Establish the MVP baseline
A current baseline report with known failure cases.
Open lesson05 · Retrieval engineering
Semantic search, keyword search, and hybrid retrieval
A retrieval-method comparison table.
Open lesson06 · Retrieval engineering
Chunking and document structure
A chunking decision record for the demo corpus.
Open lesson07 · Retrieval engineering
Metadata, filters, and access boundaries
A metadata and access-control matrix.
Open lesson08 · Retrieval engineering
Query handling and reranking
A small retrieval experiment plan with a measurable hypothesis.
Open lesson09 · Grounded answer generation
Context assembly and evidence budgets
A context-assembly diagram and evidence-budget recommendation.
Open lesson10 · Grounded answer generation
Prompts that answer only from evidence
A versioned QA prompt specification.
Open lesson11 · Grounded answer generation
Citation design and source fidelity
A citation contract for the MVP.
Open lesson12 · Grounded answer generation
Refusal, uncertainty, and hallucination control
A refusal policy and safety test cases.
Open lesson13 · Reliability & product readiness
Failure isolation and graceful degradation
A dependency failure matrix.
Open lesson14 · Reliability & product readiness
Latency, cost, and observability
An MVP performance budget and telemetry checklist.
Open lesson15 · Reliability & product readiness
Security, permissions, and deployment boundaries
An MVP threat and permission checklist.
Open lesson16 · Reliability & product readiness
Pilot review and 30-day implementation plan
A prototype readiness report and next-30-day roadmap.
Open lessonCurrent status
A completed 16-lesson course, with no enrollment currently available.
Completed: 16 lessons · 30-day MVP path. The course materials are complete and based on practical RAG engineering and evaluation work. If a learning or waitlist workflow becomes available, this page will be updated.