🎓 Free, open-source course on deterministic AI verification
Part of QWED-AI • Member of NVIDIA Inception Program
🎯 Jump to: 📺 Video Intro | 📈 Your Progress | 🚀 Choose Your Path | 🗺️ Course Map | ❓ FAQ
📈 Track Your Progress (Click to Expand)
Module Completion:
- Module 0: Prerequisites (20 min)
- Module 1: The Crisis (30 min)
- Module 1.5: Physics of Failure (45 min)
- Module 2: The Theory (45 min)
- Module 3: Hands-On (60 min)
- Module 4: Advanced (45 min)
- Module 5: Verification Landscape (45 min)
- Module 6: Domains (60 min)
- Module 7: Context Engineering (60 min)
- Module 8: Agentic Integration (60 min)
- Module 9: DevSecOps (45 min)
- Module 10: Advanced Patterns (45 min)
- Module 11: Legal Auditor (60 min)
💡 Pro Tip: Save this page (bookmark) or fork the repo to track your checkboxes!
Which role matches you? (Click to Expand)
Learning to integrate LLM verification into APIs
Module 0 → Module 1 → Module 3 (Hands-On) → Module 9 (DevSecOps)
Building verification for regulated workflows
Module 1 → Module 2 → Module 6 (Finance Domain) → Module 11 (Legal Auditor)
Master verification theory + advanced patterns
All modules + Capstone Project
Understanding neurosymbolic AI fundamentals
Module 2 → Module 1.5 → Module 5 → Module 10
⏱️ How Much Time Do You Have?
- ⚡ 30 mins: Module 1 (The Crisis)
- 🚀 2 hours: Core Developer Path (Hands-On)
- 🎓 Full Course: 8-10 Hours (Spread over 2 weeks)
| Module | Time | Focus | Best For | Difficulty |
|---|---|---|---|---|
| 0: Prerequisites | 20m | Fundamentals | New to LLMs | ⭐ Easy |
| 1: The Problem | 30m | The Problem | Everyone | ⭐ Easy |
| 2: Theory | 45m | Logic | Engineers | ⭐⭐⭐ Hard |
| 3: Hands-On | 60m | Code | Builders | ⭐⭐ Medium |
| 6: Domains | 60m | Industry | Business | ⭐ Easy |
| 11: Legal | 60m | Law | Legal Tech | ⭐⭐ Medium |
The Problem:
- Developers ship LLM-powered apps without verification
- No one teaches verification fundamentals
After This Course:
- ✅ Understand determinism vs probabilistic systems
- ✅ Implement formal verification in production
- ✅ Use mathematical proofs to catch hallucinations
- ✅ Ship provably correct AI outputs
Think of it this way:
🎨 LLMs are Artists
- Creative and convincing
- Bad at precise details
- Don't ask an artist to do your taxes!
🧮 QWED is the Accountant
- Boring and strict
- Never makes a math mistake
- This is who you want handling your money!
Visual Workflow:
graph LR
A["User Query<br/>Natural Language"] --> B["LLM Artist<br/>Creative & Fast"]
B --> C["Draft Answer<br/>May contain errors"]
C --> D["QWED Accountant<br/>Strict & Deterministic"]
D --> E{"Mathematically<br/>Proven?"}
E -->|"✅ Yes"| F["Verified Output<br/>100% Confidence"]
E -->|"❌ No"| G["Error Report<br/>+ Explanation"]
style B fill:#ffc107
style D fill:#4caf50
style F fill:#4caf50
style G fill:#f44336
Do I need a GPU?
No! You can run everything locally with:
- Ollama (free, runs on CPU)
- Or use OpenAI API (cheap for learning)
How long is this really?
- Fast track (skipping videos): 3-4 hours
- Full course with videos: 8-10 hours
- With hands-on capstone: 12-15 hours
Spread over 2-3 weeks at your pace.
Will I get certified?
GitHub doesn't issue certs, but you'll build:
- A verified banking agent (portfolio piece)
- Production-ready verification patterns
- Cryptographic Audit Trail for compliance
Module 1: The Crisis (30 mins) ✅
Why LLMs can't be trusted + Real $12,889 bug.
Module 1.5: The Physics of Failure (45 mins) ✅
Deep dive: Why LLMs hallucinate and why verification is NECESSARY.
Module 2: The Theory (45 mins) ✅
Determinism, symbolic reasoning, verification approach.
Module 3: Hands-On (60 mins) ✅
Build your first verifier with QWED + Production examples.
Module 11: The Legal Auditor (60 mins) ✅
Build a Deterministic AI Paralegal with qwed-legal.
(See "Track Your Progress" at top for full list)
Quiz: Why can't RAG alone prevent hallucinations? (Click for Answer)
Answer: RAG provides context, but it doesn't solve reasoning errors. If the retrieved document says "Revenue is $5M" and the LLM calculates "Profit = $5M - $6M = $1M", RAG can't catch that math error. Verification (like QWED) checks the logic deterministically.
By the end of this course, you will be able to add this seal of trust to your own AI agents:
By the end, you'll have:
- ✅ Verified Banking Agent that refuses to steal
- ✅ CI/CD Pipeline that blocks hallucinating PRs
- ✅ Cryptographic Audit Trail for compliance
Last Updated: January 2026 | 11 Modules | Growing Community
CC0-1.0 - Public domain. Free to use, modify, and share!