Co-Design a Self-Service Data Tool with AI
Learning Sciences for Innovators (LSFI) is a CMU Africa and Mastercard Foundation initiative that helps fellows build and scale EdTech in Africa.
As the sole designer on a data efficiency project, I leveraged AI prototyping tools to build a self-service platform that cut processing time from 6 hours to under 10 minutes.
My Role
Prompt Engineering
Data Analyst
Usability Testing
Team
Me :) + Claude 3.7 Sonnet
Timeline
Mar 2025 (2 weeks)
Tool
Lovable.ai
CONTEXT
LSFI is an initiative that empowers African EdTech fellows through learning science course certification program
PROBLEM
However, finding out how to issue the certificates has been a headache for the LSFI team
CHALLENGE
How might we empower a non-technical team to manage complex data efficiently and independently?
RESULTS
Am AI-powered data processing tool, works 36x faster than manual processing
Monthly → Daily
FINAL DESIGN
Introducing: Learner Certification Dashboard
Step 1: Bulk upload raw data files exported from LMS
Step 2: Customize certification standard, filter by date
Step 3: One click, and find out who's eligible for which certificate
PROCESS DEEP DIVE
Fast prototyping and iteration with AI
In a two-week sprint, I collected the project requirement, researched the best AI tool for rapid prototyping, implemented the tool, and iterated based on PM feedback.
How did I analyze the project needs?
UNDERSTANDING THE PAIN POINT
Shadowing to understand the problem
To design an effective solution, I first shadowed the Program Manager to understand the end-to-end certification data workflow—from file exports to final certification.
Ease of use for non-technical team members
Reduces manual work and accelerates certification turnaround
Option to adjust certification pass thresholds as needed
How did I select the right tool for long-term success?
TOOL SELECTION
Researching on implementation methodology
To start off, I experimented with a range of possible solutions to test out their capabilities and assess whether they match with our team's needs.
How did I collaborate with AI for this project?
PROMPT ENGINEERING
Approach prompt like design —
understand the user, structure the logic, and iterate for clarity and reliability
I started by mapping out the structure of both exported file from LMS, identifying what data was essential (e.g., names, emails, quiz scores) and how fields differed in format.
Instead of one large, vague prompt, I designed step-by-step instructions for the AI to navigate the complexity of data processing.
Example
I included sample data and expected outputs to help the AI visualize what I needed.
Example
I framed prompts around what the user does (e.g., uploading files, clicking “Show Results”) so the AI’s responses fit seamlessly into the intended workflow.
Example
Even with well-structured prompts, AI outputs weren’t always perfect. Just like how one would program, I added intermediate output checkpoints (e.g., “Show merged file before filtering”) to verify each step and catch the specific reason behind each unsuccessful attempt.
Additionally, I ask Lovable.ai to explain its understanding of the project, and check if it has any gap with my intended workflow.
LEARNINGS
AI is powerful—but only with the right guidance
X
Designer-AI collaboration relies on design thinking.
It takes rounds of iterating, refining, and building.
