Retrieval-Augmented Generation (RAG)
Grounding AI with real knowledge — embeddings, vector search, and citation-backed answers
Session Topics
Why RAG?
LLMs hallucinate when they don't know something. RAG replaces hallucination with retrieval — grounding answers in real, citable documents.
Embeddings
What embedding vectors are, how semantic similarity works, and why 'king - man + woman = queen' is more than a curiosity.
Semantic Search
Moving beyond keyword matching to meaning-based retrieval. Cosine similarity, nearest-neighbour search, and re-ranking.
Vector Databases
Pinecone, Weaviate, Chroma — how they store and search millions of embedding vectors at speed.
RAG Architecture
The complete retrieve-augment-generate pipeline: chunking, indexing, query embedding, retrieval, context assembly, and generation.
Build Your Resume RAG System
Build a functioning Retrieval-Augmented Generation system using your resume as the knowledge base. Answer questions about the resume with citations. Compare RAG vs non-RAG responses to see the power of grounding. This is what LinkedIn, Bing, and legal tech use.
Compare a resume against 3 job descriptions using embeddings, and present which job is the closest fit with evidence.
Key Takeaways
- RAG is the single most impactful technique for reducing LLM hallucination
- Embeddings convert meaning into geometry — similar meanings cluster together
- Chunking strategy dramatically affects retrieval quality
- RAG + structured output + prompt engineering = most enterprise AI stacks