Build your product.
Understand every
decision in it.
12 weeks. 2 hours live per week. We build your AI product together — and you leave capable of driving it without depending on anyone.
Tools you'll use in the program
A course
Theory and frameworks
Passive learning
Me building it for you
Another subscription you forget
Real product, being built weekly
Technical decisions explained in context
You executing — I guide
PRs, diagrams, real code
A portfolio you can present on Day 84
Tired of being the person
who doesn't understand
what's being built.
Product Managers
Want to drive AI decisions without depending on engineers to translate everything.
Founders
Want to understand what they're building — and stop being blocked by technical opinions.
Aspiring AI PMs
Want the technical depth that separates "product manager" from "the person engineers respect."
Builders with an idea
Want to build something real, with guidance — not alone and guessing.
Three blocks. One system.
Each week has a concept, a hands-on scope, and a deliverable you submit as a PR. You don't move forward until you can explain what you just built.
THINK - Weeks 1-4
System design, data contracts, failure modes.
Where does the AI actually fit? Map every component. Separate deterministic logic from LLM reasoning.
What does your data allow? Schema to mock dataset. You prompt the AI, the AI generates.
What goes in. What must come out. Golden examples written before any code.
What must never happen? Guardrails, edge cases, failure modes — designed before they hit production.
VALIDATE - Weeks 5-8
Prompts, structured outputs, RAG, evals, observability.
Write the first real prompt. Force the model to return structured JSON. Test against golden examples.
Build the eval dataset. Run prompt v1. Log everything. Find the failure patterns.
Give the system memory. Retrieve similar past decisions semantically from the same PostgreSQL you already use.
Define success criteria before writing the next prompt. Compare v1 vs v2 in Braintrust.
BUILD - Weeks 9-12
Backend, frontend, streaming, edge cases, demo.
Spec-Driven: you write the spec, Cursor executes. Backend with REST API + RAG + LLM + Structured Output.
Interface that streams responses word by word. The standard for any AI product in 2025.
Guardrails in code. What happens when the LLM fails, the schema is invalid, RAG finds nothing.
You present the system — architecture, decisions, every piece — to real stakeholders. Without notes.
What eval-driven development produces
Accuracy climbing from 50% to 86.67% across prompt versions — Braintrust eval pipeline. This is what Week 8 looks like in practice.
How a session works.
30 min
Concept + Diagram
I explain the week's concept with a diagram. You ask questions until it's clear.
60 min
Hands-on
You execute. I guide. One atomic scope — something real you can point to at the end.
30 min
Debrief + Decision
You explain back what you built in your own words. If you can't, we're not done.
Golden rule: If you can't explain back what you did, the session isn't over.
Day 84. What you know.
How to decompose an AI system into responsibilities
How data schema defines what your product can actually do
How I/O contracts work in production systems
How to structure, version, and improve prompts systematically
What Structured Outputs are and why free text fails in production
What RAG is and how to implement it without new infrastructure
How to evaluate an LLM system with Langfuse and Braintrust
What eval-driven development is and how to apply it as a PM
How to read and review AI-generated TypeScript code
How to write specs that AI executes (Spec-Driven)
What LLM streaming is and how it works in the interface
How to present an AI system to technical and business stakeholders
A real week. In 90 seconds.
Program walkthrough — coming soon.
I'll walk through a real week: the concept, the hands-on scope, and the deliverable.
This is happening now.
TQH Inventory AI
Los Angeles, USA - Fashion Operations
A Product Manager with zero coding background is building an AI-powered inventory decision assistant for a fashion brand — from architecture diagram to working product. Every week: a concept, a hands-on scope, a PR to the repo.
Stack: TypeScript · PostgreSQL · pgvector · OpenAI Structured Outputs · SSE streaming
Learning arc: System architecture → Data contracts → Prompts → RAG → Evals → Backend → Demo
Full case available: Day 84
The complete case — what she knew before, what she built, what she can explain on Day 84 — will be documented here.
From people in the process.
“week 7 was the one that broke my brain in a good way. i finally understand what RAG actually does. not the definition — like, WHY you'd use it. my eng stopped having to explain things twice and i think he respects me more now lmao”
Senior PM
Week 7 of 12
“honestly thought i'd be lost the whole time but pedro is very good at meeting you where you are. i pushed my first real PR on week 9. legit cried a little not gonna lie. nobody in my career ever taught me this stuff”
Founder (non-technical)
Week 10 of 12
$2,500. 12 weeks.
Real product. Real learning.
Two hours live per week. Your product. Your repo. Your PRs. I guide every decision — but you make them.
$2,500
12 weeks · fixed
What's included
12 x 2h live sessions (weekly)
Architecture diagram for your product
Private GitHub repo with weekly PRs
Prompt library and eval dataset
Working backend + frontend
Session recordings
Async support between sessions
Final demo preparation
Don't have a product idea yet? Start with Phase 0 — I'll structure the product before we start building it together.
Stop being the person
who doesn't understand
their own product.
12 weeks from now, you'll have a working AI product — and you'll be able to explain every decision in it.
© 2026 Pedro Brandão LTDA · hi@pdrobrandao.com
← Back to portfolio