Nexa AI

Model Hub is an offline integration hub where various AI models can be downloaded, as well as a community for programmers to discuss and share insights on AI.

---

Final Delievrables

(👈click here)

---

The primary challenge of this project is to streamline the user flow, helping users navigate more efficiently and reduce the drop-off rate.

I had zero knowledge about the AI programing at that time, so I started my project with a various of questions 💭.

👉🏻Who are our users?

👉🏻What are their user needs and goals?

👉🏻What is not ideal for now?

To understand the questions above, I started my research to better understand the problem space as well as our users.

I read through 8 research reports 📖 and product docs 📋, did 4 talks with different stakeholders 🙌 and finished a lot of CUJ(Critical User Journey meetings) and self-study on the AI programing domain.

---

1. Complex download process. 

Before👈 | After 👉🏻

before

after

---

2. Lack of guidance or support.

According to our data, 72% of people choose to abandon the download page for the SDK and model.

“Seems way to complicated.”

I split this page into two parts: a separate SDK download page and a simple, straightforward model download page.

Before

After

“It’s easy to get stuck and not know what to do next.”

We investigated the backend, collected the most common issues. The top issue is that users don’t realize they need to download the SDK first.

I placed the SDK download in nevi bar and added reminders next to each model in our Model Hub.

I put everything into a single guide to assist users with the download. We also offered multiple ways for them to report any issues they encounter

After the update, the SDK download drop-off rate decreased by  21%.

---

3. Hard to find idea model.

53% drop off for the model hub page

We discovered that users are less likely to filter by RAM or disk size. Instead, they prefer filtering by the model's use case.

Before

After

After the update, the model hub download drop-off rate decreased by  11%.