❄️ Winter Bootcamp · July 2026 · Day 3 · Wednesday

Inteli Camp — Day 3

Prompts from the Model — the prompt that builds the platform

🎯 What we'll do today

Turn the spec (revised in yesterday's visit) into a prompt that describes the platform, not a one-off task
Design the solution flow: spreadsheet → screening → invitation → video → decision
Define the architecture and the AI screening rule
Generate the first version in Lovable and iterate

🏁 Deliverable of the day

Platform-building prompt written and parameterized
First version generated in Lovable, with the main flow navigable
Screening rule returning structured output with no sensitive attributes

Learning objectives

What you'll be able to do by the end of the day

Objectives

✅ By the end, you can

Derive from the spec a prompt that describes a product, not an answer
Structure the building prompt: objective, users, screens, data, rules and constraints
Read an architecture diagram and map each part to the prompt
Specify AI screening with structured output
Recognize bias and privacy risks in a platform that filters people

🧭 Why it matters

In Lovable, the prompt is the platform's blueprint — the better the blueprint, the better the build
A robust platform is predictable, fair and testable
It's the bridge between the spec (Day 1) and the working MVP (Day 4)

Week Roadmap

Where we are in the journey

Day 3/5

Day 1 · Mon Spec & Modeling ✅ Spec validated, acceptance criteria set

Day 2 · Tue Technical Visit ✅ Visit done, spec revised

Day 3 · Wed Prompts & Platform Today: building prompt + 1st version in Lovable

Day 4 · Thu Sprint, Refinement & Tests Working MVP, tested and refined

Day 5 · Fri Presentations Pitch to the panel

✍️ Product Prompt 🗂️ Day 1 Spec 💜 Lovable 🔁 Test and iterate

From a request to a product

Today the prompt doesn't answer a question — it describes a system

Key shift

❌ Task prompt

Evaluate this candidate: [answer]

Solves one case, once
Takes no spreadsheet, stores nothing, notifies no one
Not a solution the HR team at Azul Linhas Aéreas uses day to day

✅ Product prompt

Build a platform that takes a spreadsheet of candidates, screens them by the role's criteria and invites those approved to record a video.

Describes screens, data, rules and integrations
Runs at scale: dozens of candidates per role
It's what Lovable turns into a real application

The shift

AI screening still exists — but now as one part of the platform, not the whole product.

From spec to building prompt

Each part of the Day 1 spec becomes a part of the platform blueprint

Bridge · from spec to prompt

Spec — Business DriverWhy the solution exists

→

Prompt — Product objectiveThe platform's purpose

Spec — ActorsWho uses it (HR, candidate)

→

Prompt — Users and screensWhat each role sees and does

Spec — Functional RequirementsWhat the solution does

→

Prompt — FeaturesUpload, screening, invite, video

Spec — EntitiesCandidate, role, evaluation

→

Prompt — Data modelDatabase fields and relations

Spec — Non-Functional RequirementsExpected quality (ISO 25010)

→

Prompt — ConstraintsLGPD, anti-bias, usability

Principle

Nothing in the platform comes from nowhere — every screen and rule traces back to a spec decision.

The screening platform

The candidate's path, from spreadsheet upload to the HR decision

Solution

📥Step 1Spreadsheet uploadHR uploads an Excel/CSV with the role's candidates

🧮Step 2AI screeningScores each candidate by the role's criteria and builds the ranking

✉️Step 3Automatic inviteApproved candidates get an email to record a presentation video

🎥Step 4Video collectionThe candidate submits the video, attached to their profile

🧑‍⚖️Step 5Human decisionThe recruiter reviews ranking and videos and decides who moves on

📌 Scope for Azul Linhas Aéreas

Instead of HR reading spreadsheet by spreadsheet, AI screening ranks candidates by fit to the role and automates the invitation to the next stage — the final decision remains with the HR team.

🏗️ Platform architecture

The layers Lovable generates from your prompt

Architecture

Users

🧑‍💼 Recruiter (HR)

Uploads the spreadsheet, follows the ranking and decides

🧑‍🎓 Candidate

Gets the email invite and submits the video

▼

Interface — Frontend generated in Lovable

📥 Upload

Imports the candidate spreadsheet

📊 Ranking panel

Candidates ordered, with score and rationale

🎥 Video gallery

Gathers the videos received per candidate

▼

Processing

🧹 Ingestion

Reads and validates the spreadsheet fields

→

🤖 Screening engine (AI)

Scores each candidate by the role's criteria and returns score + rationale

▼

Data and integrations

🗄️ Database

Candidates, roles, scores, status and videos

✉️ Email service

Sends the invite to approved candidates

🎞️ Storage

Keeps the submitted videos

Anatomy of the building prompt

Seven parts that make the AI build the right platform

Prompt · Concept

🔧 The 7 parts

Objective: what the platform solves and for whom
Users: the roles (recruiter, candidate)
Screens and flow: upload, panel, videos — and the order
Data: the entities and fields (candidate, role, score)
Business rules: how screening scores and who passes
Constraints: LGPD, anti-bias, usability (from the NFRs)
Definition of done: when the platform is acceptable

📋 Example — prompt excerpt

Build a web platform for candidate screening.
Users: recruiter and candidate.
Screens: (1) spreadsheet upload, (2) ranking panel, (3) video gallery.
Data: candidate (name, email, education, answers), role, evaluation.
Rule: score each candidate 0–5 by the role's criteria; approved ones get an email to record a video.
Constraints: do not use age, gender or origin; follow LGPD.

Screening under the hood

Inside the platform, the AI evaluates each candidate and returns structured data

AI component

🧱 Why structured output

Free text feeds neither the ranking nor the email trigger
A fixed format (JSON) becomes data: the platform sorts, filters and decides the invite
Always ask for: score per criterion + rationale + recommendation

For each candidate, reply in JSON:
{ "candidate": ..., "score": 0-5,
"rationale": ..., "recommendation": invite|review }

📊 What the panel receives

Candidate	Score	Recommendation
Ana	5	Invite to video
Bruno	4	Invite to video
Carla	2	Review — below cutoff

Trigger

Score ≥ cutoff → the platform sends the email invite automatically.

Responsible AI in screening

In a platform that filters people and sends emails, this is not optional

Ethics · NFR

⚠️ Four mandatory safeguards

Bias: screening ignores age, gender, origin and appearance — it evaluates only what relates to the role.
Privacy (LGPD): the spreadsheet carries personal data; use only what's needed and protect access.
Human decision: the AI recommends, the recruiter decides. Never an automatic rejection.
Communication: the email is an invite, not a promise — and the candidate should know AI is in the process.

🔗 Where it enters the platform

They are constraints of the building prompt — straight from the spec's NFRs (Day 1)
They connect to ISO/IEC 25010: security and reliability
Write it explicitly — in the screening rule and in data access control

Constraints: evaluate only role-related competencies; do not expose sensitive data; flag uncertainty instead of inventing.

Iterate in Lovable

The platform isn't born ready — each round fixes one part

Iteration

🔁 The cycle

Write the building prompt with the 7 parts
Generate the platform in Lovable
Navigate the flow with a sample spreadsheet
Locate what broke: missing screen, wrong field, loose rule
Adjust only that part of the prompt and generate again

🧪 Test spreadsheet

A strong candidate (should be invited)
A weak candidate (should fall below the cutoff)
A borderline candidate or with missing data
If the platform misses the edge case, the fix is in the screening rule

Bridge to Day 4

Refinement and testing happen tomorrow, in the development sprint.

Parameterization and common failures

A platform that serves many roles — and what tends to break

Robustness

🎛️ Parameterization

Mark what changes: [role], [criteria], [cutoff score]
The same platform now serves several roles at Azul Linhas Aéreas
In Lovable, each parameter becomes a configurable field
Test with different roles to ensure it stays valid

Failure	Cause	Fix in the prompt
Generic screen	Prompt didn't describe the fields	List the screens and each one's fields
No email sent	Trigger not specified	Describe: score ≥ cutoff → send invite
Inconsistent screening	No criteria or examples	Define criteria and calibrate with examples
Signs of bias	No attribute constraint	Forbid age/gender/origin in the rule

🔨 Hands on — write the prompt and generate in Lovable

Group sprint · 30-min timer · write, generate and test the flow

⏱️ Active Learning

🧭 Sprint steps

Write the building prompt with the 7 parts
Generate the platform in Lovable
Upload a spreadsheet with 3 candidates: strong, weak and borderline
Walk the flow and adjust one part at a time

👥 Group roles

Who writes the prompt (keeps the single version)
Who generates and navigates in Lovable
Who records what changed each iteration

✋ Call the professor when

The flow doesn't close after 2 adjustments
Screening shows any sign of bias
You don't know which part of the prompt to fix

🏁 Acceptance criteria

Done when

The spreadsheet imports, the ranking shows with score and rationale, and approved candidates trigger the invite — with no protected attributes used.

Checklist — is the building prompt robust?

Answer "yes" to all before considering the platform ready

Self-assessment

Does it have a clear platform objective?
Does it define the users (recruiter and candidate)?
Does it describe the screens and flow in the right order?
Does it specify the data (entities and fields)?
Does the screening rule say how to score and who passes?
Does it ask for structured output (JSON/table)?
Does it describe the email invite trigger?
Does it carry constraints for bias and privacy?
Is it parameterized for other roles?
Does it make clear the decision is human?

🎓 Day 3 wrap-up

Where we got and what comes next

Wrap-up

✅ Today you

Turned the spec into a prompt that builds the platform
Defined the architecture and the screening flow
Specified the screening rule with structured output
Generated and iterated the first version in Lovable

➡️ Next days

Day 4 (Thu): development sprint, refinement and tests of the platform
Day 5 (Fri): presentation to the panel

📌 Before tomorrow

Record the building prompt and the test spreadsheet used. Tomorrow the sprint begins: arrive with the platform generating a first version of the screening funnel.