Video summary

سریع‌ترین مسیر مهندس هوش مصنوعی شدن در ۲۰۲۶ | درآمد، مهارت‌ها و پروژه‌ها

Main summary

Key takeaways

Technology

Summary of the video (AI engineer roadmap + earning/proj tips)

Big picture / who this video is for

  • The video argues that many people learn AI skills in the wrong order, wasting time on skills companies aren’t hiring for.
  • The creator, Hooji, builds a roadmap based on:
    • Job ads (e.g., LinkedIn and major job sites)
    • Observing employees in roles similar to those at OpenAI / Anthropic / Meta-style companies
  • The goal: explain what an AI Engineer actually does, what companies want in 2026, which projects get hired, and how to earn money inside Iran.

What an “AI engineer” does (vs AI researcher)

  • AI researcher: invents a new “engine” (new models / research).
  • AI engineer: uses existing models to build a real product/application that people can use.
  • Key emphasis: most hiring is for building applications/systems using existing models, not training from scratch.

Market + compensation claims

  • Mentions ~500,000+ AI engineer job openings globally.
  • Cites an average salary of ~$242,000/year.
  • Mentions expected growth of ~26% by 2033.
  • Includes a notable claim: “40 needed skills” don’t yet exist in the job market, suggesting newcomers can get ahead by starting now.

Skills roadmap (time-boxed phases)

Phase 1: Fundamental engineering skills (≈ 1.5–3 months)

  1. Python (production-ready)

    • Not “intro tutorials”—the goal is to write code ready for real deployment.
    • Topics: data structures, functions, file handling, error handling, production quality
    • Rationale: most AI tooling/frameworks run on Python.

  2. Git & GitHub (mandatory)

    • Version control skills.
    • Specifically: repos, meaningful commits, branches, merge requests
    • Framed as your portfolio/resume.

  3. Core ML concepts

    • Understand key terms:
      • what a model is
      • training vs inference
      • what inference means
      • embeddings
    • Purpose: be able to communicate with a team and read documentation.

Phase 2: Large Language Models & practical model usage (≈ 2–3 months)

  1. Prompt engineering

    • Positioned as undervalued beyond “just chatting with GPT.”
    • Mentions techniques such as:
      • “Fujosh learning” (likely a subtitle/translation issue)
      • chain-of-thought prompts
      • output formatting for more consistent reliability

  2. APIs (OpenAI/Anthropic APIs)

    • How to send requests and handle responses.
    • Practical focus: controlling token usage and cost.

  3. Hugging Face

    • Suggested because many companies prefer running models locally (especially for sensitive data) instead of sending everything to external APIs.

Phase 3: Build complete AI systems (≈ 2–3 months)

  1. LangChain

    • Presented as a popular framework to connect models with databases and tools and build multi-step workflows.

  2. RAG (Retrieval-Augmented Generation) / “Rag To”

    • Core business problem: large models don’t inherently know a company’s internal data.
    • RAG lets the model access PDFs/docs/internal databases and answer based only on them.
    • Mentions vector databases.
    • Creator references a full RAG tutorial video (link mentioned).

  3. AI agents

    • More than chatbots: agents that do work, like sending emails, calling APIs, or updating databases.
    • Mentions having basics via another linked resource.

  4. MCP protocol

    • Described as a “new standard” from Anthropic (subtitles may show “Entropic”).
    • Goal: connect models to tools like Google Docs, GitHub, Zapier using MCP.
    • Mentions a dedicated thorough MCP link.

Phase 4: Turn skills into job-ready proof (resume-projects)

  • Advice: build projects that convince employers.
  • Build three specific projects, each including:
    • A clean GitHub repo (with README)
    • An architectural diagram
    • A demo video
  • Project ideas:
    1. Document risk/analysis system
      • Analyzes heavy documents, finds risks, and provides accurate cited sources.
    2. Natural language “data analyst” agent
      • Converts a user question into SQL, queries a database, and returns results + charts/graphs.
    3. Automated ticketing agent
      • Reads and categorizes tickets and performs actions automatically in external systems.

Earning money inside Iran (freelance + productized repeats)

  • Claims emigrating isn’t required to earn well; cites local salary context around ~90 to 140 million tomans (compared to abroad).
  • Emphasizes freelancing on Iranian and foreign sites.
  • Example income range: ~20 to 400 million tomans per project, for things like:
    • store chatbots for image processing
    • speech recognition
    • automation of company tasks
  • Key strategy: build once, reuse/repackage
    • After the first project (example: ~2 months, ~300 tomans), similar projects can be completed in 2–3 weeks for similar or higher earnings.

Startup / fixed salary option (also mentioned)

  • Notes Iranian startups in fintech, education, healthcare may need AI experts and may offer:
    • fixed salary + project percentage
  • Creator preference: choose the best-income option by working on the same project.

Key takeaway

  • Long-lasting foundations highlighted: Python + RAG.
  • Encouragement: start before being “fully ready”—learn by building and iterating through mistakes.

Main speakers / sources

  • Speaker: Hooji (creator/host)
  • Primary external sources mentioned:
    • Job ads on LinkedIn and major job sites
    • Observations of employees at OpenAI and Anthropic
  • Tools referenced:
    • LangChain
    • Hugging Face
    • MCP (Anthropic)
    • OpenAI/Anthropic APIs
    • GitHub
    • Zapier

Original video