Six AI Engineering Skills, Six Portfolio Projects to Practice Them

Bashiri Smith · Facebook reel · 2026-09-25 · 0:19 · 139,540 views · Open on Facebook

Topics: Portfolio Projects, Start Here: Roadmaps & Strategy · Level: intermediate

Summary

A quick list that pairs six core AI engineering skills (RAG, agents, LLMOps, evals, fine-tuning, safety) with one hands-on project for each. Each project goes beyond a basic demo and covers something that matters in production, like multi-hop retrieval, durable agent state, provider failover, calibrated judges, distillation and red-teaming on every deploy. The creator offers a longer breakdown, plus 20–21 more projects, to people who comment "SKILLS".

Key points

• RAG: combine a knowledge graph with vector search (graph + vector hybrid) so multi-hop questions get answered correctly.
• Agents: build a multi-agent research assistant with a supervisor agent coordinating workers and durable (persisted) state.
• LLMOps: build a self-healing LLM gateway that automatically fails over to another provider when one goes down.
• Evals: implement LLM-as-judge and calibrate it against real human labels so you can trust its scores.
• Fine-tuning: distill a task from a frontier model into an 8B model to cut inference cost sharply.
• Safety: build an automated red-team harness that runs on every deploy, like a regression test for safety.
• The video only names the projects. No specific tools, frameworks or models are recommended.

Resources mentioned

• Bashiri Smith's AI Engineer Projects Breakdown · pdf · check price
  The creator's step-by-step guide to building the six projects in the video, plus about 20 more project ideas.
  DM only: comment SKILLS on this reel on Facebook and you get it by DM.

Try this

• Pick the skill you want to practice and build the matching project.
• Comment "SKILLS" on the reel to get the creator's full breakdown and the extra project ideas.
• RAG: a hybrid knowledge graph + vector retrieval system that handles multi-hop questions.
• Agents: a multi-agent research assistant with a supervisor agent and durable state.
• LLMOps: a self-healing LLM gateway that fails over when a provider goes down.
• Evals: an LLM-as-judge evaluator calibrated against real human labels.
• Fine-tuning: distill a frontier model's task performance into an 8B model at a fraction of the cost.
• Safety: an automated red-team harness that runs on every deploy.