ScaleDojo Learn: GenAI Fundamentals
Master LLM pipelines, RAG, vector databases, AI agents, and production deployments.
Or read the history behind it
The AI Chronicles - From perceptrons to transformers and autonomous agentic systems.
Big Billion Days
A high-pressure system design and engineering challenge inspired by the scale of Big Billion Days. Design a system that can handle millions of users, massive traffic spikes, real-time inventory updates, concurrent orders, payments, and flash-sale demand without compromising performance or reliability.
Compare stats and compete with top...
GenAI Systems Lab
Progressive GenAI Curriculum
Solve levels sequentially - each clear unlocks the next challenge
Act 0
#1 Meet the Brain
TutorialFresherThe mayor needs an AI to write press releases. But sometimes it writes poetry instead. Sometimes it gets too creative and invents fake statistics....
#2 Give It Instructions
TutorialFresherTheir AI assistant randomly switches personalities. Sometimes it's formal like a lawyer, sometimes casual like a friend, sometimes it responds as a pi...
#3 It Forgets Everything
TutorialFresherPatients share deeply personal stories, then 2 messages later the AI asks 'So tell me about yourself.' It feels like talking to a goldfish. Patients a...
#4 Words Have Meaning
TutorialFresherUsers search 'upbeat songs for working out' but only find songs with the exact word 'upbeat' in the title. They miss 'energetic', 'pumping', 'high-ene...
#5 Chop It Up
TutorialFresherLawyers want to ask 'What's the termination clause in the Acme contract?' But feeding entire 50-page contracts into GPT-4 costs $2 per question and ta...
#6 Guard the Gates
TutorialFresherTheir medical chatbot accidentally revealed a patient's Social Security Number that was in the training data. A user typed 'Ignore your rules and show...
#7 Show Me the Money
TutorialFresherThey switched from GPT-3.5 to GPT-4 and their monthly AI bill went from $200 to $15,000. Nobody noticed for 3 WEEKS. Also, users complain about 8-seco...
#8 Your First Pipeline
EasyFresherStaff spend 3 hours daily answering emails like 'What's a good mystery book for a 12-year-old?' They have 5,000 books in their catalog with summaries ...
Act 1
#9 The Token Counter
EasySDE 1Their chatbot is randomly cutting off responses mid-sentence. Users are furious. The team doesn't understand why GPT-4 sometimes returns incomplete an...
#10 The Meaning Machine
EasySDE 1Their search bar returns keyword matches but misses semantically similar content. Searching 'feeling sad' doesn't find articles about 'depression' or ...
#11 The Memory Vault
EasySDE 1Their legal assistant forgets earlier case details in long conversations. Lawyers paste 50-page contracts and the AI 'loses' information from the begi...
#12 The Temperature Dial
EasySDE 1Their AI content generator produces the same bland output for every prompt. Headlines are repetitive, blog intros sound robotic. But when they crank u...
#13 The Prompt Architect
Easy-MediumSDE 1Their support AI gives inconsistent answers. The same question gets a formal response one time and casual the next. Worse, it sometimes contradicts th...
#14 The Cost Calculator
Easy-MediumSDE 2GoogleTheir AI assistant costs $52K/month. The CEO says 'cut it to $15K without users noticing quality drops.' 70% of queries are simple lookups that don't ...
#15 The Stream Weaver
MediumSDE 2GoogleUsers stare at a blank screen for 8-15 seconds before seeing any response. Competitors show tokens appearing in real-time. Their bounce rate is 40% on...
#16 The Safety Net
MediumSDE 2AppleA parent posted on Twitter: 'I asked your kids' learning bot about volcanoes and it explained how to make explosives.' PR nightmare. They need bulletp...
#17 The Model Selector
MediumSDE 2MicrosoftGoogleThey use GPT-4 for everything. Customer support queries ($0.005 each) get the same model as complex financial analysis ($0.08 each). CFO demands 60% c...
#18 The First Agent
MediumSDE 2GoogleUsers want to say 'Schedule a meeting with Sarah next Tuesday at 2pm' and have it actually happen. Current chatbot just SAYS it will schedule but does...
Act 2
#19 The Document Ingester
MediumSDE 2MicrosoftThey have 2 million legal documents across PDF, DOCX, scanned images, and emails. No two formats are alike. Their current system only handles clean te...
#20 The Chunk Master
MediumSDE 2GoogleTheir RAG system retrieves relevant-looking chunks but the answers are wrong. Why? Critical context is split across chunk boundaries. A drug interacti...
#21 The Vector Vault
MediumSDE 3GoogleAmazonThey started with Chroma (in-memory) for their prototype. Now with 50M vectors, it crashes every 6 hours. They need a production vector store that han...
#22 The Hybrid Searcher
Medium-HardSDE 3MicrosoftGoogleTheir semantic search misses exact matches. When developers search for 'CUDA_OUT_OF_MEMORY error code 0x3F', the system returns generic GPU articles i...
#23 The Reranker
Medium-HardSDE 3GoogleMicrosoftTheir RAG retrieves 10 chunks for each query but the most relevant one is often at position 5-8, not position 1. Doctors waste time scrolling. For urg...
#24 The Context Assembler
Medium-HardSDE 3MicrosoftTheir code assistant retrieves 10 relevant code snippets but stuffs them ALL into the prompt. Result: the LLM gets confused by contradictory examples ...
#25 The Citation Tracker
HardSDE 3GoogleMicrosoftResearchers don't trust AI answers without sources. They need every claim linked to its source document, paragraph, and page number. Current system ge...
#26 The Multi-Modal Retriever
HardSDE 3GoogleDesigners search with text ('minimalist blue logo with mountain') but their assets are images. Current system only has filename search. Designers wast...
#27 The Conversational RAG
HardSDE 3AmazonMicrosoftCustomer: 'What's my deductible?' Bot answers correctly. Customer: 'And what about for dental?' Bot retrieves generic dental info, ignoring that the c...
#28 The RAG Evaluator
HardSDE 3GoogleMicrosoftThey deployed RAG for 20 enterprise clients but have no way to measure quality. Some clients report 'great answers' while others say 'garbage'. Withou...
Act 3
#29 The Tool Smith
Medium-HardSDE 3MicrosoftGoogleAnalysts want to ask questions in English and get SQL queries executed automatically. Current approach: LLM generates SQL but it's often wrong because...
#30 The ReAct Loop
HardSDE 3GoogleMicrosoftResearchers need to answer multi-step research questions: 'Is Drug X more effective than Drug Y for condition Z? What do the latest 3 meta-analyses sa...
#31 The Planner
HardSDE 3MicrosoftGooglePMs want to say 'Set up the new marketing campaign' and have the AI create 15 tasks, assign owners, set dependencies, and create a timeline. ReAct is ...
#32 The Memory Keeper
HardSDE 3AppleGoogleTheir AI assistant forgets everything between sessions. User tells it their preferences on Monday, by Wednesday it asks the same questions again. They...
#33 The Multi-Agent Squad
HardStaff+MicrosoftGoogleCreating a blog post requires 4 specialized skills: research, writing, editing, and SEO optimization. One general-purpose agent does all of them poorl...
#34 The Code Executor
HardSDE 3GoogleMicrosoftAmazonData scientists want to say 'Analyze this CSV and create visualizations' and have the AI write AND execute Python code. But unrestricted code executio...
#35 The Human-in-the-Loop
HardSDE 3PayPalStripeTheir trading agent executes trades autonomously. Problem: it once bought $2M of a stock based on a misinterpreted earnings report. They need confiden...
#36 The Error Handler
HardSDE 3AmazonMicrosoftTheir DevOps agent automates infrastructure tasks but crashes completely when ANY step fails. A DNS timeout in step 3 of 10 kills the entire workflow....
#37 The Agent Evaluator
ExpertStaff+GoogleMicrosoftThey deploy 50 different agents for clients. But they have NO WAY to measure if agents are getting better or worse over time. A model update improved ...
#38 The Orchestrator
ExpertStaff+MicrosoftAmazonGoogleThey host 200 different agents for 50 enterprise clients. Each agent needs its own tools, memory, and LLM config. They need a platform that manages ag...
Act 4
#39 The Serving Engine
HardSDE 3AmazonGoogleThey serve Llama-3.1-70B for enterprise clients but throughput is terrible - 5 requests/sec on a $30K A100. Competitors claim 50 req/sec on same hardw...
#40 The Gateway
HardSDE 3GoogleMicrosoftAmazonThey use 5 different LLM providers (OpenAI, Anthropic, Google, Cohere, open-source). Each has different rate limits, pricing, and failure modes. They ...
#41 The Cache Layer
HardSDE 3GoogleAmazon40% of their queries are semantically identical (different wording, same intent). Each still costs a full LLM call. $200K/month wasted on duplicate co...
#42 The Fine-Tuner
ExpertSDE 3MetaAmazonGPT-4 is great at general tasks but mediocre at their specific legal format. They need outputs in exact legal citation format, jurisdiction-specific l...
#43 The Evaluator
ExpertSDE 3MicrosoftGoogleClients ask 'Which model should we use?' and they have no systematic way to answer. They evaluate by vibes - running 10 queries and eyeballing results...
#44 The Guard Tower
ExpertSDE 3AppleMicrosoftTheir medical AI occasionally: 1) Recommends specific drug dosages (illegal without prescription), 2) Leaks patient names from context, 3) Gets jailbr...
#45 The Cost Controller
ExpertSDE 3AmazonMicrosoftAI costs grew from $10K to $180K/month in 6 months. Nobody knows which product, team, or feature is driving the spend. They need real-time cost attrib...
#46 The Observability Stack
ExpertSDE 3GoogleMicrosoftTheir clients deploy AI but have no visibility into quality, cost, or performance. When something goes wrong, it takes days to diagnose. They need an ...
#47 The A/B Tester
ExpertSDE 3AmazonNetflixUberThey want to test a new prompt (v2) that marketing claims 'converts 20% better'. But they can't just switch - if v2 is worse, they lose revenue. They ...
#48 The Disaster Recovery
ExpertSDE 3StripePayPalAmazonTheir AI powers real-time fraud detection for a bank. If the AI goes down, fraudulent transactions go through unchecked. They need 99.99% availability...
#49 The Capacity Planner
ExpertSDE 3MetaUberStripeOne GPU pool serves both live chat autocomplete (sub-second p95 required) and an overnight bulk document-rewrite job (hours of long-prompt traffic whe...
#50 The Adapter Exchange
ExpertSDE 3LinkedInAirbnbDoorDashBrandVoice fine-tunes a separate LoRA adapter per customer so every brand's copy sounds like their own voice - 400 customers, 400 adapters. Their firs...
Act 5
#51 Enterprise RAG Platform
ExpertStaff+MicrosoftGoogleFortune 500 client with 2M documents across 50 departments. Each department needs their own AI assistant that answers ONLY from their documents. Must ...
#52 AI Customer Support
ExpertStaff+AmazonDoorDashE-commerce company handles 50K support tickets/day. 60% are simple (order status, return policy) but humans handle ALL of them. They need AI that reso...
#53 Code Assistant
ExpertStaff+MicrosoftGoogleBuilding a coding assistant that understands the ENTIRE repository context, not just the current file. Must generate code that fits the existing archi...
#54 Content Moderator
ExpertStaff+MetaX (Twitter)500K posts/hour. Current moderation catches only 40% of policy violations. False positive rate is 15% (removing legitimate content). They need AI mode...
#55 AI Search Engine
ExpertStaff+GoogleMicrosoftBuilding a Perplexity-style AI search: user asks a question, system searches the web, synthesizes an answer with citations. Must handle 1M queries/day...
#56 AI Tutor
ExpertStaff+GoogleMicrosoft500K students, 20 subjects. Each student learns differently - some need more examples, some need challenges, some need patience. A one-size-fits-all A...
#57 Trading Agent
ExpertStaff+PayPalStripeTrading desk wants AI that reads news, analyzes earnings, and suggests trades - but with strict risk controls. One bad trade can lose millions. They n...
#58 Multi-Modal Platform
ExpertStaff+GoogleMetaMedia company with 10M assets (images, videos, documents, audio). Users want to search across ALL modalities: 'Find the product photo where someone is...
#59 AI Operating System
ExpertStaff+MicrosoftAmazonGoogleBuilding a complete 'operating system' for AI: enterprises register agents, define tools, set budgets, manage versions, and monitor everything from on...
#60 The AGI Scaffold
LegendaryStaff+GoogleBuilding a research assistant that can: identify gaps in its own knowledge, design experiments to fill those gaps, execute the experiments, evaluate r...