How AI Transformed My Data Design Journey

December 31, 2024

Designing an intuitive interface for integrating diverse data sources into a catalog sounds straightforward, right? Well, it’s not. Especially when you’re dealing with the chaos of e-commerce data—product databases, customer APIs, sales transactions, and inventory files—all needing to come together seamlessly.

The challenge wasn’t just about building something functional; it was about making it simple enough for a data analyst with minimal technical expertise yet robust enough for seasoned data scientists to extract value.

I turned to AI—Llama 3.2 (3.2B), a local Large Language Model (LLM), to help me navigate this journey. What unfolded was a fascinating collaboration between human creativity and machine intelligence that made this daunting task not only manageable but genuinely exciting.

Task overview

Here’s what I was working on:

  • Objective: Design a user-friendly interface to connect various data sources, cleanse the data, and load it into a catalog, specifically for e-commerce applications.
  • Target Users: Data Analysts and Data Scientists in e-commerce, with varying technical expertise.
  • Core Challenge: Simplifying complex workflows—like connecting to databases, APIs, and files—into an experience that’s intuitive and efficient.

Step 1: Setting up my AI partner

Before diving into the design process, I needed to set the stage. This meant installing Ollama, downloading Llama 3.2, and setting up Open WebUI to interact with the model.

Once everything was ready, I crafted a custom system prompt to guide Llama 3.2:

dataset-catalog

I asked it to simulate expert roles—like a developer, data analyst, or data scientist—offering advice tailored to my needs. The goal was to create a collaborative environment where we could brainstorm, explore scenarios, and evaluate trade-offs like two colleagues working on the same problem.

This setup turned Llama 3.2 into more than just a tool; it felt like having a highly skilled teammate who always had fresh ideas to offer.

Step 2: Figuring out the user flow

Once the AI was ready, we got straight to work. The first task was to define the user flow—mapping out how someone would go from connecting a data source to saving cleansed data in a catalog.

Here’s how it played out:

dataset-catalog

Llama 3.2 provided a clear, step-by-step user flow, emphasizing key interactions like error handling, authentication prompts, and simplifying repetitive tasks.

Follow-Up: I refined these suggestions, prompting it for more details specific to e-commerce (like handling real-time sales APIs or bulk inventory uploads).

dataset-catalog

By the end, I had a robust flowchart that balanced simplicity and functionality.

Step 3: Wireframing with AI’s help

With the user flow in place, I started sketching wireframes. Here’s where Llama 3.2 truly shined—it turned into a brainstorming powerhouse.

Connect from database

dataset-catalog

Wireframe:

dataset-catalog

Suggested clear, concise forms for connecting to databases, with example fields like hostnames, credentials, and schemas.

Connect from APIs

dataset-catalog

Wireframe:

dataset-catalog

Seamless API integration, focusing on inputs like API endpoint, authentication method (API key, OAuth), and request parameters.

Connect from Files

dataset-catalog

Wireframe:

dataset-catalog

Connect Files wireframe that supports seamless CSV/JSONL uploads, with clear input fields for file selection, schema mapping, and optional metadata

Iterated on these outputs, tweaking the wireframes to fit the flow and user needs.

Dataset catalog

dataset-catalog

Wireframe:

dataset-catalog

Dataset catalog showcasing diverse data sources with intuitive metadata organization, cleansing status, and seamless interaction for users.

Dataset cleansing

dataset-catalog

Wireframe:

dataset-catalog

Interface with intuitive tools for handling missing values, removing duplicates, and applying simple transformations, guided by AI-driven insights.

Step 4: Bringing it all together

Finally, I compiled everything—flows, wireframes, and cleansing tools—into a cohesive dashboard design.

Dashboard

The platform simplifies connections, enabling users to effortlessly link databases, APIs, or files without feeling overwhelmed by complexity. Additionally, the catalog views are thoughtfully organized, presenting metadata in a clean and intuitive manner that enhances usability and clarity.

dataset-catalog

Dataset cleansing

Data cleansing was a critical piece of this puzzle. It’s one thing to connect a dataset but making it usable is where the real magic happens.

dataset-catalog

This phase felt less like “designing” and more like sculpting with a co-pilot constantly nudging me toward better solutions.

Why AI was a Game-Changer

This wasn’t just a design project; it was a collaborative experiment. AI wasn’t here to replace me—it was here to amplify my creativity, challenge my assumptions, and speed up iterations.

What made AI indispensable?

  • Expert Simulations: Llama 3.2’s ability to mimic different perspectives (developer, analyst, scientist) added depth to the design.
  • Quick Iterations: I could test ideas and get feedback in seconds, which would have taken days otherwise.
  • Clarity in Complexity: Its structured responses helped me untangle complex workflows and focus on user experience.

Final thoughts

What started as a daunting task of designing a Data Catalog Integration and Cleansing platform turned into a deeply rewarding experience, thanks to the power of AI. Tools like Llama 3.2 are more than just assistants—they’re creative partners that push you to think bigger, faster, and smarter.

This project reinforced something I’ve always believed: the best designs happen when technology and human intuition work hand in hand. And in this case, AI proved to be an incredible teammate.