Navi | Devpost

Inspiration

During a previous internship, we realized how difficult it is for AI labs to source high-quality training data for CUA (Computer Use Agents). Collecting this type of behavioral data is expensive, time-consuming, and often fragmented, leaving many promising research ideas under-resourced.

What it does

Navi is a gamified platform for gathering supervised training data on real-world computer usage. Users are given entertaining, goal-oriented tasks — for example, “Add an item to your Amazon cart and proceed to checkout.” While completing these tasks, they allow Navi to securely capture on-screen activity such as mouse clicks, navigation flows, and (where permitted) keystrokes.
All data is validated for accuracy and quality, then anonymized and delivered to organizations developing productivity assistants, next-generation AI systems, and CUA (Computer Use Agents).

How we built it

We developed a Next.js frontend that handles task management, data ingestion, validation, and secure storage. The platform’s architecture makes it easy to extend features like task marketplaces, leaderboards, and client dashboards.

Accomplishments that we’re proud of

We’re proud that Navi bridges a significant gap for the AI and CUA communities: it empowers labs and companies to train the next generation of Computer Use Agents with richer, more authentic data while giving individuals an opportunity to contribute — and benefit — from their everyday computer use.

What we learned

We discovered an enormous and growing demand for unique, high-fidelity training data, especially for human-computer interaction and CUA development. Supply hasn’t kept up with that demand, and projects like Navi can help close that gap by aligning user incentives with research needs.

What’s next for Navi

Our next step is to expand into specialized datasets — from productivity workflows to niche applications like creative tools, developer environments, and educational software — helping to accelerate the development of foundational AI models and Computer Use Agents across industries.