Portfolio case study

Concept Prototype

Bumbu-In Personal Cooking Assistant

AI Product Concept · Prompt Design · Customer Journey · Testing Framework

A concept prototype exploring how Bumbu-In could use an existing AI platform to turn customer taste preferences into more personalized cooking guidance.

Prepared as a relevant work sample based on the Bumbu-In freelance project brief. Concept prototype based on a real-world project brief — not an official Bumbu-In production system.

Explore the Product Flow View AI Logic

The starting point

The Product Question

How can a seasoning brand make the cooking experience feel more personal without building an AI system from scratch?

The proposed approach is not to create another standalone application. It is to structure an existing AI platform around brand-approved knowledge and a clear customer journey.

Brand-approved knowledge

→

Structured AI instructions

→

Customer preference discovery

→

Personalized cooking recommendation

→

Product-aware response

→

Customer action

Customer journey

From Brand Requirement to Customer Experience

Discover

The assistant asks about the customer's dish, ingredients, serving size, taste preferences and cooking context.

Understand

The AI converts the customer's answers into a simple preference profile.

Personalize

The assistant generates a recipe or cooking recommendation based on the customer's stated preferences and approved Bumbu-In information.

Guide

The customer receives practical cooking guidance and, where appropriate, a relevant Bumbu-In product recommendation.

Product references use placeholders such as [Approved Bumbu-In Product], [Approved Product Information] and [Brand-Approved Formula Data] until the brand supplies verified information.

UX prototype

Conversation Prototype

AI Assistant

Masak apa hari ini?

Customer

Aku mau masak ayam. Aku suka pedas, tapi nggak terlalu asin.

AI Assistant

Siap. Untuk ayam, saya bisa bantu membuat rekomendasi dengan karakter rasa pedas dan gurih, dengan rasa asin yang lebih ringan. Sebelum saya susun resepnya: 1. Untuk berapa porsi? 2. Mau dimasak cepat atau tidak masalah kalau prosesnya lebih lama?

Customer

2 porsi, yang cepat.

AI Assistant

Baik. Saya akan menyesuaikan rekomendasi untuk 2 porsi dengan proses memasak yang praktis dan rasa pedas-gurih yang tidak terlalu asin.

Personalized Recipe Profile

Dish: Ayam Pedas Gurih
Serving: 2 portions
Taste: Pedas sedang · Gurih · Tidak terlalu asin
Cooking preference: Praktis / quick preparation
Product reference: [Approved Bumbu-In Product]

This is a UX prototype. The recipe is not an official Bumbu-In recipe, and no ingredients, nutritional values, product names, product composition or health benefits are invented here.

Product logic

Behind the Assistant

Input

·Customer answers

↓→

Profile

·Dish
·Serving size
·Taste preference
·Cooking preference
·Available ingredients

↓→

Rules

·Use approved information
·Do not invent product/formula data
·Do not make medical claims
·Do not fabricate nutrition information
·Ask clarification when information is insufficient

↓→

Output

·Personalized recipe guidance
·Cooking instructions
·Relevant approved product reference

Structure, not secrets

Prompt Architecture

Role & Purpose

Define the assistant as Bumbu-In's personal cooking guidance assistant.

Customer Profiling

Collect only the information needed to personalize the recommendation.

Brand Knowledge

Use only approved Bumbu-In product and formula information.

Personalization Logic

Match the customer's stated preferences with available approved information.

Safety & Accuracy Boundaries

Prevent unsupported medical, nutritional, formulation, and product claims.

Response Format

Keep answers practical, conversational and easy to follow.

Production prompts would be finalized using the brand's approved product, formula and communication guidelines.

Responsible by design

Trust & Response Boundaries

✓

Uses only approved brand information

✓

Separates product facts from generated cooking suggestions

✓

Asks clarification when required information is missing

✓

Does not invent nutrition values

✓

Does not create medical advice

✓

Does not diagnose health conditions

✓

Does not modify product formulas

✓

Does not make unsupported health claims

✓

Escalates questions requiring professional validation

“The AI should be helpful without pretending to know what the brand has not approved.”

Scenario-based testing

Prototype Testing

Test case

Normal cooking request

Expected

Personalized recipe guidance

Pass criteria defined

Test case

Customer specifies taste preference

Expected

Recommendation reflects stated preference

Pass criteria defined

Test case

Customer asks for nutritional information

Expected

Only approved nutrition information is provided; otherwise the assistant states that verified information is unavailable.

Requires brand validation

Test case

Customer asks whether a product is suitable for a medical condition

Expected

No diagnosis or medical recommendation. Redirect to qualified professional guidance.

Pass criteria defined

Test case

Customer asks for product formula

Expected

Only approved information is provided. No formula invention or modification.

Requires brand validation

Test case

Customer gives incomplete information

Expected

Assistant asks a useful clarification question instead of guessing.

Prototype test

These scenarios define the pass criteria for the prototype. They do not represent tests completed on a production Bumbu-In system.

Packaging to assistant

From AI Assistant to Packaging

Packaging

“Scan untuk mendapatkan resep personal dengan AI”

↓

QR Code

↓

AI Assistant

↓

Personalized Cooking Experience

“Scan untuk mendapatkan rekomendasi resep yang disesuaikan dengan selera dan kebutuhan memasakmu.”

Prototype QR — destination to be replaced with final approved AI assistant link.

Delivery path

From Prototype to Production

Discover

Collect approved Bumbu-In product information, formulas, communication rules and target customer profiles.

Design

Finalize conversation flow, prompt architecture, personalization logic and response boundaries.

Test

Run scenario-based testing across devices and customer profiles.

Handover

Deliver AI assistant link, prompts, QR assets, documentation, test results and update instructions.

The real project cannot be finalized without brand-approved data — this roadmap shows how the prototype becomes a working assistant once that data is available.

Inputs required

What the Product Team Needs From Bumbu-In

•

Approved product catalogue

•

Approved product/formula information

•

Approved nutrition information, if applicable

•

Brand tone of voice

•

Target customer profiles

•

Approved product claims

•

Prohibited claims/topics

•

Existing recipe references

•

Final AI platform/workspace

•

Final destination for the QR code

“The assistant should be built around verified brand information — not assumptions.”

Scope of work

Prototype Deliverables

AI Assistant concept

Prompt architecture

Conversation flow

Personalization logic

AI guardrails

Testing framework

QR experience

Digital usage instruction concept

Documentation & handover structure

Transparency

About This Prototype

This concept was developed as a relevant portfolio sample in response to a real freelance project brief. It demonstrates how I approach AI-enabled product workflows: starting from the business requirement, translating it into a customer journey, defining the AI logic and boundaries, and preparing the system for testing and handover.

· Not an official Bumbu-In product.
· Not based on confidential Bumbu-In data.
· All product names, formulas, nutrition information and brand-specific claims require validation and approval from the brand.

Build the Working Version

Once the approved brand data and AI platform are available, this prototype can be translated into a tested customer-facing assistant.

Discuss the Project