AI Engineer Mistakes Pt. 3: Don't Grind Math First, Choose Embeddings Carefully

Bashiri Smith · Facebook reel · 2026-09-12 · 0:31 · 13,812 views · Open on Facebook

Topics: Start Here: Roadmaps & Strategy, Embeddings & Vector Databases, Retrieval-Augmented Generation (RAG) · Level: beginner

Summary

This short video contrasts a "wannabe" with a $200K+ AI engineer across two common mistakes. First, don't hold off on building until you've finished studying linear algebra and calculus. Learn the math when a problem actually needs it. Second, don't assume all embedding models are the same. Pick one suited to your domain, rerank what retrieval returns, and measure recall before you trust it. The video ends by pointing to the creator's free roadmap guide.

Key points

• Studying all the math (linear algebra, calculus) before building with AI is a mistake. Don't treat the math as a permission slip you need before starting.
• On the job, AI engineers aren't asked to derive backpropagation by hand. Learn the math when a problem demands it.
• Embedding models are not interchangeable. The wrong one can push the correct answer down to 'page two' of your retrieval results.
• Choose an embedding model suited to (or tuned for) your domain instead of using the default.
• Rerank the retrieved results before passing them on.
• Measure retrieval recall before you trust the pipeline.

Resources mentioned

• Bashiri Smith's AI Engineer Roadmap (free resource guide) · pdf · free
  The creator's free guide of resources for becoming an AI engineer, sent to viewers who comment "ROADMAP".
  DM only: comment ROADMAP on this reel on Facebook and you get it by DM.

Try this

• Start building with AI now; learn the math only when a problem needs it.
• Don't use the default embedding model without checking it. Pick one suited to your domain.
• Add a reranking step to your retrieval pipeline.
• Measure retrieval recall before trusting results.
• Comment "ROADMAP" on the video to get the creator's free resource guide.