Solo designer and researcher. Minerva University, 2023. I owned every stage: user interview, probe design, two rounds of prototyping, deployment, analysis, and design direction.

How do you understand someone's relationship to their environment and wellbeing when they can't easily put it into words? Interviews tell you what people think they feel. I needed something that could capture what they actually feel, in context, over time.

I designed a physical probe kit: a set of self-guided activities my participant could complete over several days in their own space. Things like photographing what feels like an obstacle, mapping a room's emotional experience with color, or playing a card game with friends. Each activity was meant to externalize an internal state without asking for self-analysis.

The kit went through two prototyping rounds. After self-testing the first version, I redesigned activities that felt too academic for someone in their early twenties. I made them more playful and open-ended to lower the barrier to participation.

The interview told me she was analytically minded, future-anxious, motivated by progress. The probe told me something different: her sense of wellbeing was mostly social. Her academic obligations were her "obstacle." Her flatmates were her "gratitude." The only activity she didn't enjoy was the only one she did alone. She texted me after the final activity (a social card game) to say she'd really liked it.

Two product ideas came out of this: a gamified social check-in tool (like We're Not Really Strangers but for daily wellbeing) or a lightweight platform where close friends share struggles and wins. Both come from the same place: for this user, wellbeing is social, not individual.

Field researcher, information architect, content designer. 6-person team with The Open Box Project (architecture collective), Hyderabad, India, 2023. I led field research, routing decisions, the mapping workshop, and content strategy. Visual design by Anoushka Shome (Open Box).

Hyderabad has 1,000-year-old monuments but no usable tourist guide for English speakers. Existing ones were either overwhelming (50 sites, no logic) or shallow (nice photos, no practical info). The real problem was spatial: the major sites sit in two clusters several kilometres apart. Any guide that ignored that would give tourists exhausting, inefficient days.

Research first
We walked every route before designing anything. Visited 12+ sites, verified opening hours, timed visits, and scouted food stops along the way.

Information architecture
We ran a mapping workshop to solve the two-cluster problem. We prototyped groupings by geography, theme, and walking distance, and landed on three color-coded day routes. The color system let tourists orient without reading everything first.

Contextual content placement
Food recommendations appear at the point in the route where you'd actually be hungry, not in a separate appendix. Nimrah Cafe appears right after Charminar because it's a one-minute walk. This required walking the route, not just knowing the city.

A printed physical guide and a companion digital version linked by QR code. The final product combined two perspectives: local architects with cultural depth, and international students with the outsider's confusion (which is closer to the actual tourist experience).

Solo designer-researcher. Minerva University, 2023. 2.5-week sprint. I ran field observation, expert interviews, a participatory workshop, AI-assisted prototyping, and wrote the final design proposal.

Of Hyderabad's 9,100 km of roads, only 500 km had usable footpaths. On the ground, I saw thin painted lines, sandy margins, and walking space blocked by parked cars and vendors. Everyone walked in the road. The deeper issue was that no one had defined what a sidewalk in Hyderabad should actually do, so even new construction missed shade, seating, accessibility, and social space.

Field research
Walked two routes with specific observation goals, photographed and mapped barriers and workarounds.

Expert interview
Spoke with local architect Saagar Tulshan to understand why pedestrianization stalled despite budget.

Co-creation workshop
Recruited locals from a nearby coffee shop. Two exercises: map what exists, map what you wish existed. People valued trees, food stalls, culture. They wanted shade, seating, quiet, water, lighting.

AI-assisted prototyping
Generated 100 sidewalk renderings using getimg.ai to explore design directions faster than I could sketch. Shared with Mumbai architect Anoushka Shome for critique. Her feedback was that the images showed aesthetics, not inclusive construction. So I pivoted and turned the deliverable from visual renderings into a written design proposal for engineers.

A values-based design proposal organized around three principles: Disability Inclusion, Physical/Mental Wellbeing, and Social/Cultural Health, each with specific recommendations.

Limitations worth naming: the disability section draws from ADA/UN standards rather than conversations with disabled Hyderabadis. The workshop sample was small. The proposal was never reviewed by engineers. But the process taught me to start with the people who use a space, not the people who build it.

Senior capstone project, Minerva University, 2025. Traditional methods for identifying transit deserts rely on census data and transit schedules. These are quantitative measures that miss what it actually feels like to navigate a street. Are there sidewalks? Shelters? Is the infrastructure maintained or falling apart?

I collected 50,000+ Street View images from Houston via Mapillary, then used Mask2Former (a deep learning segmentation model) to extract visual features like sidewalks, transit stops, pedestrian pathways, and road infrastructure. I merged those image-derived features with American Community Survey census data and existing transit access metrics to build a more context-aware classification of transit deserts.

The street-level imagery told a different story than the traditional models. Some census tracts that looked fine on paper had huge variation block to block. One side with maintained sidewalks, the other with nothing. Tract-level analysis was too coarse to catch that. The things surveys miss (crumbling sidewalks, missing shelters, hostile infrastructure) turned out to matter a lot for accurate classification.

The question behind this project is one I keep coming back to: how do you get detailed, on-the-ground information at scale? Surveys cover a lot of ground but miss what things actually look like. Walking a neighborhood gives you that context, but you can't walk every neighborhood. Computer vision let me split the difference: process 50,000 street scenes and pull out details that census data just doesn't capture.