Learning path map
Detailed content
📊 The Frameworks Worth Their Weight in Gold
The six frameworks the Factory uses under the hood: maturity, quick wins, build vs. buy, ROI, roadmap, and governance. They make the output look like—and deliver the value of—BCG and KPMG work.
The BCG maturity curve: Passive → Aware → Active → Operational → Transformational. You place the company at one of the 5 stages and score 7 dimensions from 1 to 5.
It’s the first deliverable: it shows where the company is today. Without it, everything else (the roadmap, ROI) floats in midair.
BCG maturity curve · 5 stages · 7 dimensions · readiness radar.
AI initiatives that run in <60 days, cost <US$50k, carry low risk, and have demonstrable ROI. Prioritized in an effort × impact matrix.
They’re the quick proof of value that unlocks budget. This is the framework for your Mission in this module.
Low effort × high impact · <60 days · demonstrable ROI · momentum.
A weighted decision matrix (time, 3-year cost, customization, internal capacity, maintenance, risk) that indicates whether the company should build or buy for each use case.
It’s the million-dollar question in every AI project. A clear framework prevents decisions based on guesswork.
Weighted matrix · 3-year TCO · vendor comparison · vendor lock-in.
Four sources of value (efficiency, revenue, risk reduction, strategic value) become simple ROI, payback, and 3-year NPV with a 10% discount rate.
It’s the deliverable that convinces the finance team. Numbers speak louder than a pretty slide.
4 sources of value · simple ROI · payback · NPV · sensitivity analysis.
Five time-based phases: Foundation (30) → Quick Wins (60) → Scale (90) → Optimize (180) → Transform (360), each with goals, milestones, and owners.
It’s what turns "use AI" into a plan the client opens on Monday and starts executing.
5 phases · checkbox milestones · owner per activity · leading vs. lagging indicators.
Data classification (Public → Restricted), what can and cannot become AI training/input/output, roles (CDO, stewards), and compliance (LGPD, GDPR, CCPA).
It’s the deliverable that puts legal and the board at ease. Without governance, no pilot makes it to production.
Data classification · acceptable use policy · data stewards · LGPD/GDPR.
🗜️ Lazy RAG: huge PDF → cheat sheet
McKinsey, BCG, and KPMG playbooks are 28–46 pages long—they blow past any context limit. The solution is lazy RAG: split them into chunks, summarize each one, and create an ultra-compressed cheat sheet.
Dumping an entire 46-page PDF into Claude Code exceeds the context limit—and fills the window with useless buzzwords. You need the valuable information, not the volume.
Understanding the bottleneck is what justifies the technique. Without that, you waste expensive context on junk.
Context limit · signal vs. noise · token cost · value density.
Extract the text from the PDF, count tokens with the tokenizer, and split it into chunks of ~10.000 tokens. Each chunk fits comfortably in an API call.
It’s the heart of lazy RAG: process one part at a time instead of everything at once.
10k-token chunk · tokenizer (cl100k_base) · PyMuPDF · process in parts.
A prompt that asks only for frameworks, statistics, actionable insights, citations, and terms — preserving numbers and cutting buzzwords. A block with no value gets the [MINIMAL_CONTENT] label.
The prompt is where the cheat sheet's quality comes from. A bad prompt = a generic, useless summary.
Structured extraction · preserve numbers · cut buzzwords · empty content marker.
The model that synthesizes: huge context, rock-bottom price, with exponential retries and delays between calls to respect the rate limit.
It’s what makes it possible to distill an entire PDF library for pennies. The right model in the right place.
Gemini 2.5 Flash · low cost · exponential retry · rate limit.
Combine the synthesized blocks in order, discard those marked as empty, and generate a single Markdown file with a header, sections, and a "check against the original" footer.
It’s the output that goes to Claude Code. Well put together, it becomes a reusable reference; poorly put together, it becomes noise.
Concatenate by index · discard empty values · header + footer · final Markdown.
The actual script: reads a folder of PDFs, saves progress in progress.json, skips blocks already completed and supports --dry-run e --single-pdf.
Just ask Claude Code to run it. Resuming ensures you don’t pay twice for the same block.
ProgressTracker · resumable · --dry-run · skip completed blocks.
🧠 Your consultant brain
Standalone cheat sheets aren't enough—they become powerful when organized into a knowledge base that Claude Code can consult on demand. Here you build your personal consultant library: curated, versioned, and ready to inject.
Generating text is cheap; space in the context window is expensive. Distilling means trading volume for density: every token that goes in needs to earn its relevance.
It’s the mindset that separates those who fill the context with junk from those who inject only gold. It defines the quality of everything.
Distillation · information density · context efficiency · signal vs. volume.
Choose authoritative sources: the free guides McKinsey, BCG, KPMG, OpenAI, and IBM publish to showcase their expertise. You get elite knowledge for free.
The quality of the foundation depends on the quality of the source. Garbage in, garbage out—choose authoritative sources.
Authoritative sources · free guides · curation · borrowed authority.
The fixed structure of a good cheat sheet: concepts & frameworks, critical statistics, actionable insights, quotes, and terminology. Always easy to scan.
Consistent structure is what makes the foundation searchable and injectable. Each leaf becomes a modular piece.
Fixed sections · scannability · bolded term · 1 page per source.
Organize the folder conhecimento/ by topic, date each cheat sheet, record the source, and version it with git. The knowledge base evolves without becoming a mess.
Curated knowledge is an asset that grows in value over time—if you don’t lose control of it.
Folders by topic · date + source · version with git · knowledge base as an asset.
Instead of dumping the entire knowledge base, you pull only the cheat sheet relevant to the current task. Lazy RAG: the human does the retrieval on demand.
It’s what keeps the window lean and the response sharp. Right context > more context.
Lazy RAG · manual retrieval · on demand · lean context window.
The folder conhecimento/ consolidated: several curated cheat sheets that power the entire Factory and go with you from project to project.
It’s the brain that makes the Factory sound like an elite consulting firm. Built once, used with every client.
Reusable library · feeds the Factory · portable asset · compounding advantage.