Aspiring Software Engineer/AI Engineer
I'm a self-taught developer passionate about Python, AI/ML, and building projects that solve real-world problems. I enjoy learning through hands-on projects, exploring new technologies, and sharing what I learn along the way. Currently exploring AI engineering, full-stack development, and open-source projects.
Markdown supported · ⌘↵ to post
The Gen-Z framing makes this approachable, but one nuance that's worth adding is that RAG isn't really a "memory" system it's a retrieval system. The quality of the final answer depends far more on retrieval quality than most people realise. If the wrong chunks are retrieved, or relevant context never makes it into the prompt, even the best model can produce a confident but incorrect answer.
In production systems, teams often spend more time improving chunking strategies, metadata, hybrid search (keyword + vector), and reranking than swapping to a larger LLM. Better retrieval usually delivers a bigger accuracy gain than a bigger model because the model can only reason over the context it actually receives.
Techlte World
The "open-book exam" analogy nails it. People genuinely think AI has some kind of persistent memory, when really it's just retrieving relevant chunks and reasoning over them in the moment. The embeddings + vector search breakdown was also super easy to follow, especially the Spotify comparison.