GenAI Systems Lab: An Interactive Lab for AI Engineers to Experience Production System Failures in the Browser

This open-source project provides a zero-backend, pure-frontend interactive learning platform for AI engineers and product managers. Users can configure RAG systems, observe failures, understand root causes, and explore over 100 production system modules (including RAG failure simulation, reasoning optimization, Agent loops, evaluation design, MCP protocol, etc.) directly in the browser. The core philosophy is: "Configure systems, observe failures, understand reasons."

In production, AI systems often fail unexpectedly. For example, a RAG system with top_k=1, no re-ranker, and a "helpful answer" strategy might retrieve outdated docs (3-year-old policy) and give wrong answers—only discovered when customers complain. GenAI Systems Lab solves this by letting engineers experience these failures before deployment, learning why they happen and how to prevent them.

This is a fully open-source, zero-backend, pure-static platform. No registration, no backend services, completely free. Tech stack: React18 + Vite6 + Tailwind CSS v4, deployed on Vercel free tier, using localStorage for client-side state persistence. Key design choices: Zero backend (scripted scenarios are more reliable than real-time API failures), local storage (no cross-device sync, no GDPR issues), PWA support (installable, offline use).

1. Concepts: 11 modules (Tokenizer, Embeddings, Context Window, Agent Loop, Guardrails, Debug RAG, etc.) with "HowTo-first" design. 2. RAG Lab:5 production failure scenarios (stale retrieval, hallucination, prompt injection, context overflow, multi-hop failure) with graded scoring. 3. Agent Modules:7 topics (ReAct Pattern, Tool Use + MCP Protocol, Agent Memory, etc.). 4. Systems:16 deep modules (Evals, Model Strategy, Cost/Latency, Fine-Tuning, etc.). 5. Playground:5 hands-on challenges (injection attack simulation, chunking comparison, etc.) +30 prompt library.

Credibility Labels: Each module has labels (✓ Mathematically faithful, ~ Simplified, ◌ Conceptual) to ensure transparency. PrepLab Modes:3 learning modes—Assessment (timed exams with skill analysis), Trainer (instant feedback + voice input + weak spot tracking), JD Prep (JD keyword extraction, skill gap analysis, targeted practice). Comparison: Outperforms YouTube/blogs, DeepLearning.AI, fast.ai in production failure simulation, JD-aware prep, zero backend, etc.

New AI Engineer: Quickly learn RAG failure modes in 1 hour (instead of months in production). Interview Prep: Use JD Prep mode to identify skill gaps (e.g., Agent design, reasoning optimization) and get targeted practice. Team Training: Weekly assessment mode to track team skill progress across 8 categories.

Local Run: Clone the repo → git clone https://github.com/SidharthKriplani/genai-systems-lab → cd into folder → npm install → npm run dev (access at localhost:5173). Online Access: Visit https://genai-systems-lab-ivory.vercel.app directly (no registration needed).