I’m Aashutosh Vyas, a high school student in the Bay Area. Most of what I work on sits somewhere between AI, software, research, and photography.

A lot of my projects start with a question I think is worth testing. Recently that has included how models can search through ideas more creatively, how human attention changes over time, how agents should work with people when they reach something they cannot solve alone, and how AI can actually be useful inside real organizations instead of becoming another chatbot layered onto an existing process.

I’m currently building Rift, working on AI implementation at SMEED CPA, and developing a few other projects across AI and software.

Outside of that, I’ve been doing photography since 2020. I mainly photograph landscapes, cities, architecture, and places that I find visually interesting. My work has been recognized through the Scholastic Art & Writing Awards, and I published my first photography book, What the Eye Returns To.

I’m usually most interested in problems where the solution is not already obvious.

I identify areas where AI can save time, remove repetitive work, or improve an existing process, then help design and implement the systems around them.

A large part is understanding how the existing workflow functions, deciding what should be automated, and building something reliable enough to become part of the production process.

Locora is a local information platform built around making useful city information easier to find.

I work across the product and engineering side, including the web platform, data systems, internal tools, infrastructure, and the technical work needed to support direct partnerships with cities.

We are currently working with multiple cities through formal partnerships.

Rift started from a problem I saw in how attention is modeled.

Most systems analyzing video treat each moment mostly independently, but human attention has memory. What someone notices now is influenced by what they just saw.

I’m building models that use temporal context to better understand how attention changes through short-form video, with the larger goal of turning that understanding into useful tools for creators and content teams.

The research has involved gaze prediction, temporal modeling, cognitive event boundaries, and experiments around what actually changes or resets the memory guiding attention.

Rift models how human attention changes over time while someone watches video.

The central idea is that attention should not be predicted frame by frame as if every moment exists independently. What someone notices now is affected by what they have already seen.

I’m building and testing models with temporal memory to capture that behavior more accurately and studying what kinds of events cause that memory to change.

Locora is a local information platform built around making useful city information easier to find.

I work across the product and engineering side, including the web platform, mobile app, data systems, internal tools, infrastructure, and the technical work needed to support direct partnerships with cities.

We are currently working with multiple cities through formal partnerships.

Emberline turns satellite thermal observations into persistent, explainable possible-fire events. It filters known industrial heat, checks official incident data, and adds weather and air-quality context for places people care about.

The app distinguishes satellite hotspots from confirmed fires. It is an informational tool, not an evacuation forecast or replacement for official emergency alerts.

Atlas is a hands-free computing system designed around people with limited upper-limb mobility.

The project explores how computer vision and alternative inputs can be combined to create a usable way to control a computer without depending on a traditional keyboard and mouse.

Atlas placed second at Grizzly Hacks. The screenshot shows the actual computer-assistant interface in its offline idle state.

Echoes is a voice conversation project with a simple web interface for starting a call.

Duo Fold explores local-first human verification for iPhone Duo. The native app asks a person to match three hinge angles, then returns the result to a small web interface.

We placed at YC Bitrig Hacks. The image is a frame from our live demo there.

This is an experimental prototype, not a secure CAPTCHA service for protecting accounts or payments.

Starline curates high-school internships, programs, competitions, and practical application guidance, helping students spend less time searching and more time moving toward their next opportunity.

Access starts with Google or email sign-in. The preview shows the actual Starline sign-in screen.

I’m working on a research question around whether an LLM can make better creative decisions by changing how it searches through ideas.

When a model is solving a creative task, there are different useful directions it can take. It can deepen the current idea, explore something much more distant, or consolidate ideas it has already found.

My approach is to represent those as explicit actions and study whether a small controller can learn when each one is most useful.

I’m especially interested in whether useful, non-obvious connections can be searched for systematically instead of being treated as something that only appears by chance.

A major part of the research is also defining how creativity should actually be evaluated. Novelty alone is not enough if the resulting idea is not useful, so the work looks at creativity through quality, diversity, and novelty rather than collapsing everything into one score.

After Passage is a photography project centered around places after people have moved through them.

The photographs focus less on people themselves and more on what remains: structures, roads, objects, landscapes, and small signs of previous human presence.

The project is about what a place can communicate even when the people who shaped it are no longer visible.

• Gold Key
• Silver Key
• 2x Honorable Mention

Recognized through the California Bay Area regional Scholastic Art & Writing Awards for photography.

I’ve built projects at hackathons across AI, accessibility, agent systems, computer vision, and human-computer interaction.

The individual projects include their specific event and placement where applicable.