AI Power User: 0 → 1
Use your gap months to gain skills and a portfolio before college. Build real AI projects — not another “click through tools” class.
Build. Break. Fix. Ship.
8 weeks. 14 modules. Real projects. Real users. A portfolio that shows you can deliver.
“Learn tools and build projects.”
Solve real problems with AI systems that actually work under constraints.
Every module follows 4 non-negotiable steps
Skip any one and you don't pass. That's how we make sure you actually learn, not just consume.
Ship something real every module. No toy demos.
Stress-test it. Find where AI quietly fails.
Improve reliability with constraints and evals.
Prove someone actually wants what you built.
14 modules. Each one builds a skill.
From first reality check through shipped tools — practice that holds up when things break.
2-Hour Reality Check
Destroy naive thinking early. What LLMs can and can't do, why 80% of AI apps fail.
Build a Daily AI Brief → break it → fix with constraints.
AI Ecosystem (Rebuilt)
Closed vs open models. API vs local. Latency / cost / quality triangle. Decision frameworks.
Same task → 3 models → justify your stack like an engineer.
How LLMs Actually Work
Token prediction, hallucinations, context poisoning, instruction hierarchy, determinism vs randomness.
Write a prompt that fails at high temp, succeeds at low.
Prompt Engineering (Upgraded)
Prompt compression, defensive prompting, strict JSON schemas. No CoT romance.
Build a Bulletproof Prompt that passes 20 test cases.
Evaluation & Reliability
Golden datasets, pass/fail scoring, edge cases, adversarial prompts. Stop building demos, start building systems.
30 test cases → score outputs → improve until ≥85% pass rate.
APIs Done Right
Retries, timeouts, fallback models, cost tracking, not just calling an LLM.
Reliable LLM Pipeline with fallback + full logging.
RAG (For Real)
Where RAG breaks, chunking strategies, retrieval errors, measurement.
Build RAG, measure retrieval accuracy, then improve it.
AI System Design
Pipeline vs agent. Stateless vs stateful. Caching. Cost vs latency tradeoffs.
Design a research assistant. Justify every component.
Agents — When You Need Them
Agents are expensive, slow, often unnecessary. LangGraph, CrewAI, and when NOT to use them.
Build an agent → replace it with a simpler pipeline.
Local LLMs
When local actually makes sense: privacy, cost at scale. Hands-on with Ollama.
Run a local model. Compare vs API on real tasks.
Build & Ship (Hardened)
Lovable.dev, Replit, Cursor. Not 'just build'. Must have users, feedback, iteration.
Ship a tool that solves a real problem with 3+ external users.
AI Product Thinking
Problem > solution. ROI thinking. Why most AI startups fail.
Interview 5 users → kill one idea → refine another.
Production Mindset
Monitoring, logging, cost dashboards, failure handling.
Stress test your system. Fix the breaking points.
Capstone Project
RAG OR pipeline + evaluation system + real users + measurable success metric.
Live app + case study: problem, design, failures, improvements.
To complete this course, you must have:
You finish with shipped work — not just notes from watching lectures.
3 shipped tools
1 with real users
1 evaluated properly
1 improved after failure
Learn from people who actually build
Aditya Karki
AI Systems & Engineering Lead
Sagar Gautam
AI Product & Build Mentor
Ready to join?
Message us on WhatsApp for cohort dates, payment, and what to prepare. Or use the form — we reply within a day. Seats are limited per cohort.
Talk to us · we'll help you- ✓Rs. 1500 total · 8 weeks
- ✓Online + Offline (Baneshwor, Kathmandu)
- ✓Real projects, not toy demos
- ✓Mentorship from working AI builders
Prefer the form? Tell us you want AI 0 → 1 and we'll follow up.