Home Programs LLMs & Chatbots
💬 LLM Program · Year-end Exhibition

LLMs & Chatbots — talk to AI, build with it, shape the future

Your child stops being a passive ChatGPT user and becomes an LLM builder. From prompt engineering to RAG, agents, and custom assistants — they ship real products. Bilingual classes (English + Persian) for Iranian families worldwide.

📚 4 seasons ⏱ 110 hours/year 🧠 Real APIs (OpenAI, Anthropic, open-source) 🎪 Year-end exhibition — ship your own AI product
✨
Open enrollment — Autumn season starts soon Free 90-minute trial class for new students.
LLM illustration
Choose your child's age band — the LLM path adapts to their level.

For under-10 learners, LLMs are introduced as friendly creative partners. Your child writes stories with AI, makes a simple "ask-me-anything" character, and learns to think critically about what AI says — a vital skill for the rest of their life.

One-year path — elementary track

Four seasons. Each season: 24 class hours + 1 workshop. Total ~110 hours/year.

🍂
Season 1 — Autumn

Meet your AI friend

24 class hours + 1 workshop

First conversations with AI through our safe, kid-friendly chat interface. Your child learns what an LLM is and how to ask great questions.

What they learn
  • 🎯 Working with Novin Zehn's proprietary IDE (kid mode)
  • What is an LLM? Why does it sometimes get things wrong?
  • Asking clear, specific questions
  • Making AI tell stories about their favorite topics
  • Safety: never share personal info with AI
📖
End of Season 1 A short illustrated story written with their AI buddy + Season 1 certificate.
❄️
Season 2 — Winter

Build your first chatbot character

24 class hours + 1 workshop

Using block-based tools and a simple drag-and-drop prompt builder, your child designs their own AI character — a wizard, an astronaut, a chef.

What they learn
  • Persona prompting (give the AI a personality)
  • Setting rules: "always be kind", "answer in Persian sometimes"
  • Block-based prompt builder (Scratch-style)
  • Voice input/output — talk to your bot
  • Testing your character with classmates
🧙
End of Season 2 A personalized chatbot character shareable with family + Season 2 certificate.
🌸
Season 3 — Spring

AI + creativity combos

24 class hours + 1 workshop

Mixing LLMs with image and voice tools. Your child makes audio stories, illustrated picture books, and short animated scenes — all co-created with AI.

What they learn
  • Combining text AI with image AI (kid-safe models)
  • Designing simple multi-page picture books
  • Recording AI-voiced narrations
  • Story-prompting in English + Persian (bilingual)
  • Critical thinking: spotting AI mistakes
🎨
End of Season 3 An illustrated bilingual storybook + Season 3 certificate.
🎪 Exhibition
🏆
Season 4 — Summer

Final project + year-end exhibition

Bootcamp + online exhibition

Your child picks a passion topic and builds a small AI app around it — an animal encyclopedia bot, a homework-helper, a bedtime-story machine. Presented at our online exhibition.

What they create
  • Final project planning + storyboarding
  • Building their AI app step-by-step
  • Recording a 2-minute demo video
  • Online exhibition with families worldwide
  • Diaspora classmates vote on favorites
🥇
End of year Annual diploma + portfolio link + showcase in our diaspora family newsletter.

For 10–14 learners, the program shifts into building real apps. Your child uses Python in our proprietary IDE, calls real LLM APIs (OpenAI, Anthropic, open-source via Ollama), and ships chatbots, study assistants, and content tools.

One-year path — middle school track

Four seasons of real Python + real APIs.

🍂
Season 1 — Autumn

Python + first API call

24 class hours + 1 workshop

Python warm-up + first OpenAI/Anthropic API call from inside Novin Zehn's IDE.

What they learn
  • 🎯 Working with Novin Zehn's proprietary IDE
  • Python refresh: variables, functions, requests
  • API keys + safety, environment variables
  • First LLM call — chat completion
  • Tokens, temperature, top-p — what they actually do
💬
End of Season 1 A working CLI chatbot + Season 1 certificate.
❄️
Season 2 — Winter

Prompt engineering + structured outputs

24 class hours + 1 workshop

Real prompt engineering. System prompts, few-shot examples, JSON outputs, role-based personas.

What they learn
  • System vs. user vs. assistant roles
  • Few-shot prompting + chain-of-thought
  • Structured JSON outputs (function calling)
  • Memory: conversation history, context windows
  • Mini-project: bilingual homework explainer
📘
End of Season 2 Personal study assistant + Season 2 certificate.
🌸
Season 3 — Spring

RAG + simple agents

24 class hours + 1 workshop

Retrieval-Augmented Generation: feed the AI your own documents. Plus first agents that use tools.

What they learn
  • Embeddings + vector databases (Chroma / FAISS)
  • Building a RAG pipeline from scratch
  • Tool calling — let the AI search the web, do math
  • Simple agents with LangChain or plain Python
  • Streamlit / Gradio web UI
🔎
End of Season 3 A web app that answers questions about uploaded PDFs + Season 3 certificate.
🎪 Exhibition
🏆
Season 4 — Summer

Capstone + year-end exhibition

Bootcamp + online exhibition

Each student ships a real LLM app: chatbot, study tool, content generator, or assistant. Presented to families worldwide at our exhibition.

What they create
  • Capstone project planning + scope
  • Front-end with Streamlit / simple HTML
  • Deployment to a public URL
  • 3-minute demo video + GitHub repo
  • Online exhibition — votes from diaspora classmates
🥇
End of year Annual diploma + portfolio + public live demo link.

For 14–19 learners, the program goes deep. By year-end your teen builds production-grade LLM apps with agents, fine-tuning, multi-modal models, and real evaluation pipelines — the same stack used at frontier AI labs.

One-year path — high school track

Four seasons. Portfolio-grade output suitable for top university applications worldwide.

🍂
Season 1 — Autumn

Transformers + prompt engineering at scale

24 class hours + 1 workshop

How transformers actually work (attention, tokens, embeddings). Pro prompt engineering. Working with multiple providers.

What they learn
  • 🎯 Working with Novin Zehn's proprietary IDE
  • Transformer architecture — attention, MLP, layer norm
  • Tokenization — BPE, vocabulary tricks
  • OpenAI, Anthropic, Mistral, open-source via Ollama
  • Cost/latency tradeoffs — choosing the right model
⚙️
End of Season 1 Multi-provider chat app with cost tracking + technical report.
❄️
Season 2 — Winter

RAG, agents, MCP, tool use

24 class hours + 1 workshop

Production-grade RAG. Agent architectures (ReAct, planner-executor). Model Context Protocol (MCP). Vector DBs at scale.

What they learn
  • Advanced RAG: hybrid search, re-ranking, eval
  • Vector DBs: Pinecone, Weaviate, Qdrant
  • Agents: LangGraph, plan-and-execute, ReAct
  • Model Context Protocol (MCP) for tool use
  • Multi-agent coordination
🤝
End of Season 2 Working multi-tool agent + technical writeup.
🌸
Season 3 — Spring

Fine-tuning, evals, multi-modal

24 class hours + 1 workshop

Fine-tune small open-source models (LoRA, QLoRA). Build real eval pipelines. Work with vision + audio models.

What they learn
  • LoRA / QLoRA fine-tuning on Hugging Face
  • Dataset prep + curation
  • Eval harnesses (LLM-as-judge, structured tests)
  • Multi-modal: vision (GPT-4V, Claude) + audio (Whisper)
  • Safety: hallucination detection, jailbreak defense
🧪
End of Season 3 A fine-tuned domain-specific model + eval report.
🎪 Exhibition
🏆
Season 4 — Summer

Capstone product + year-end exhibition

Bootcamp + global online exhibition

Ship a real, deployed LLM product. Many alumni use this capstone in their Stanford, MIT, ETH, Toronto, TU Munich applications.

What they create
  • Production capstone: SaaS, mobile, agent, or research
  • Cloud deployment (Vercel, Modal, Hugging Face Spaces)
  • Public live demo + GitHub repo
  • Technical writeup + 5-minute video
  • Exhibition: international panel of judges
🥇
End of year Annual diploma + portfolio + globally shareable capstone.
🧠
🇩🇪 Munich
Diaspora alumni stories

From bilingual classroom to a frontier AI internship

Yalda joined our online classes from Munich at 14. She started in Python, moved into AI/ML, and finished her LLM capstone with a bilingual study-tutor agent. Today she interns at a leading AI startup — and still drops by our classes on Saturdays.

"I came in just wanting to talk to ChatGPT in Persian. I left building agents I couldn't have imagined. The bilingual classes meant I never lost my mother tongue while learning the world's most cutting-edge skill."

🏅 LLM program certificates

Your child receives an official Novin Zehn certificate at the end of every season. The annual diploma + portfolio is shareable with schools, summer programs, and university applications worldwide.

1Autumn certificate
2Winter certificate
3Spring certificate
4Annual diploma + capstone portfolio
📈 Beyond year one

Two paths forward

After the first year, your child can deepen into research-grade LLM work, or branch into related areas like computer vision, AI/ML, or full-stack AI products.

🚀

Path 1 — LLM expert track

Year two: deeper into agents, MCP, fine-tuning, research papers, and shipping a real startup-grade product.

 🔭

Path 2 — Branch into related AI

Switch to AI/ML, Computer Vision, or full-stack AI app development — all building on your LLM foundation.

📞 Book your free trial class

A teacher chats with your child for 90 minutes, recommends the right age band, and answers every question you have about curriculum, scheduling, and how the bilingual classes work for Iranian families abroad.

Book Free Trial Request a Consultation
Common questions

What parents abroad ask us

Does my child need a powerful computer or GPU?

No. All hands-on work runs in Novin Zehn's cloud IDE, accessible from any laptop with a browser. Even Chromebooks work.

Are LLM API costs included?

Yes. Each student gets a class-wide API budget for OpenAI, Anthropic, and open-source models. No extra cost to families.

What about safety and screen time?

Classes are live with a teacher present. Younger learners use kid-safe wrappers around AI. We never use student data for training. Parents get weekly progress notes.

What timezones do you serve?

EST, PST, GMT, CET, AEST — we run multiple weekly time slots. All sessions are recorded for makeup.

Can my child join mid-year?

Yes. Each season has a 2-week onboarding window. A teacher will brief them on what they missed.