Human–AI Interaction Designer

I design for people who run complicated tools.

I'm Sonakshi. I trained as an ML engineer, worked as a product manager, and now design products where people meet technical systems: drone mapping, brain imaging and AI.

About

I sit between the people who build technical systems and the people who have to use them.

I studied computer science at PES University in Bangalore, where I built ML pipelines and co-authored an IEEE paper on cyclone prediction. Then I moved toward the people side: product management at a drone-mapping company and product design at a neuroimaging startup.

I'm now doing a master's in Human-Computer Interaction at the University of Maryland, where I'm a Grand Challenges Graduate Scholar in the AI and Knowledge Community cohort. I grew up between Pennsylvania and Bangalore, which taught me early that the same thing rarely reads the same way to two audiences.

User researchUsability testingInteraction designInformation designProduct requirementsUATMachine learningFigmaAdobe InDesign

Have a complicated tool? Let's make it kinder to use.

I'm open to internships and collaborations in product design, UX research and human–AI interaction.

UX research and UAT testing for a geospatial platform

Improving a geospatial platform through workflow research, observation and testing.

Role
Product management intern
Users
GIS analysts
Methods
Interviews, observation, UAT
Scale
11 research sessions, 350 test cases

Confidentiality note. To respect company confidentiality, this case study focuses on my research methods, processes and learnings. No proprietary company data, internal documents or confidential product information has been included. Any visuals or examples are recreated for illustrative purposes.

01

Mapping the GIS analyst workflow

The challenge

GIS analysts were moving from legacy geospatial tools to the company's new platform. Their work involved complex, multi-step processes, so the product team needed a clearer picture of the end-to-end workflow.

Goal

Understand the analysts' end-to-end workflow and the role each tool and output played within it.

My approach

I ran user research sessions using a master–apprentice approach, observing analysts as they worked and asking questions in context. I introduced myself as a PM intern who wanted to learn how drones are used to create orthomosaics. The analysts walked me through the process, and I asked clarifying questions wherever I got stuck.

The questions I asked, grouped into learning the basics, walking the workflow and finding decision points
Questions I asked during the sessions.

Outcome

The research produced a clearer picture of the analysts' workflow, dependencies and decision points. It gave the product team a stronger foundation for designing the new platform around real user processes.

The workflow

I mapped the workflow from drone data collection through photogrammetric processing, point cloud generation, DSMs, orthomosaics and GIS analysis.

Drone flying survey lines over terrain1 Data collectionField operators fly drones over a survey area and capture overlapping aerial images.
Ground control points on terrain2 Positioning & reference dataImage geotags are refined using PPK data, while Ground Control Points provide surveyed reference locations.
Two overlapping photos with matched points3 Photogrammetric processingOverlapping images are aligned using feature matching and reference data.
Point cloud colored by elevation4 Point cloud generationThe imagery becomes a dense 3D point cloud of terrain and visible objects.
Surface model colored by elevation5 Surface modelingThe point cloud generates outputs such as a Digital Surface Model (DSM).
Aerial map with scale bar and north arrow6 Orthomosaic generationImages are orthorectified and combined into a geographically accurate aerial map.
Map with measurements and verified checkpoints7 GIS analysis & quality checksAnalysts review outputs, verify accuracy, take measurements and prepare deliverables.
The seven stages of the GIS analyst workflow.
02

Observing real-world product usage

The challenge

As GIS analysts began using the new platform, the product team needed to understand how it fit into existing workflows, what encouraged adoption, and where the experience created friction during real tasks.

Goal

Observe analysts using the platform to identify usability issues, workarounds and points of confusion.

My approach

I conducted multiple observation sessions in the GIS lab while analysts completed their day-to-day work. I focused on understanding:

  • How analysts incorporated the new platform into their existing workflows.
  • When and why they switched back to legacy tools.
  • Where they hesitated, experienced friction or encountered bugs.
  • Which platform features they relied on most.

I documented usability issues and bugs as they surfaced and communicated technical problems directly to the engineering team.

Key findings

1. Familiarity influenced adoption

Some analysts were hesitant to move because they were comfortable with their existing tools, even when the new platform offered similar capabilities. With active project deadlines, they had not had time to try it.

2. Simplicity was a strength

Analysts who had adopted the platform were happy with its simplicity, especially the processing interface.

3. Hybrid workflows remained necessary

Analysts switched between the new platform and older tools, relying on legacy tools for capabilities that were still in development.

4. Specific issues affected efficiency

The sessions surfaced bottlenecks, including bugs I documented and reported to engineering immediately, and the four opportunities shown here.

A hand-drawn area outline next to one that follows the hill's elevationVolume calculationUsers wanted the area of interest to be selected automatically based on elevation levels.
A button being clicked repeatedly with no responseInterface responsivenessSlow response to clicks interrupted interactions.
Two charts, one with similar line colors and one with distinct colorsElevation profileGraph colors were too similar. Users wanted more vibrant colors to tell data series apart.
An export dialog with a requested scale selectorMap downloadsUsers wanted to choose the zoom scale of exported maps.
Usability issues observed, recreated as illustrations.

Outcome

The sessions showed that platform adoption depended on more than feature availability. Familiarity, project deadlines, usability and workflow compatibility also shaped analysts' willingness to transition.

I documented the issues identified and communicated bugs to engineering, helping the team understand where users experienced friction during real work.

03

Strengthening quality through UAT testing

The challenge

The analytics platform was being developed rapidly, which meant new features frequently introduced bugs and edge cases. These needed to be caught in the UAT environment before changes were pushed to production.

The team needed a more consistent way to validate functionality before release and to avoid high-priority issues surfacing late in the release cycle.

Goal

Create a structured UAT process to systematically test core workflows, document issues, and help the team prioritize bugs based on their impact on users and day-to-day work.

My approach

I helped establish a systematic testing process built around two components: a test-case sheet for finding issues, and a priority matrix for triaging them.

1. Test-case repository

I created and maintained a structured test-case sheet to document:

  • Features and workflows being tested.
  • Test steps and expected behavior.
  • Actual results and pass/fail status.
  • Bugs, reproduction steps and supporting evidence.
  • Issue status and follow-up testing.

This gave the product and engineering teams a consistent way to validate functionality, track test coverage and follow issues across releases.

The test-case sheet format with scenario, test number, test case, steps, input data, expected behaviour, result and comments columns
The test-case sheet format I created, shown with example content.

2. Prioritization matrix

I used an urgency–importance matrix to separate critical failures from lower-impact usability issues and enhancements. It gave the team a simple framework for discussing which issues needed immediate attention and which could be addressed later.

Priority matrix: P0 very high, fix now; P1 high, plan next release; P2 medium, schedule by impact; P3 low, backlog
The bug prioritization matrix.
A loop: test, log bug, prioritize, fix, retest, then release when tests pass
How the sheet and the matrix fit into each release.

Outcome

The structured UAT process helped catch high-priority issues earlier, improved visibility into test coverage, and gave the team a clearer, more consistent way to triage bugs.

It also reduced last-minute pressure on engineering before release by making issues easier to document, reproduce and address.

Clinical brain-mapping reports and product requirements

Learning a clinical neuroimaging domain, designing brain-mapping reports, and defining product requirements for a clinical neuroimaging platform.

Role
Product design intern
Users
Radiologists and hospital teams
Tools
MRIcroGL, MRView, Adobe InDesign
Scale
26+ screens specified
A clinical report layout beside a brain viewer showing tractography and network activity

Confidentiality note. To respect company confidentiality, this case study focuses on my methods, processes and learnings. No proprietary company data, internal documents or confidential product information has been included. Any visuals or examples are recreated for illustrative purposes.

01

Building neuroimaging domain understanding

The challenge

I was designing for a highly specialized clinical neuroimaging product. Before making design decisions, I needed to understand what the brain maps, tractography and other outputs actually represented.

Goal

Build enough understanding of the neuroimaging workflow and its outputs to design for clinical users, without treating the data as generic interface content.

My approach

I worked directly with fMRI-based neuroimaging data, using MRIcroGL and MRView to inspect brain maps and tractography. Through this hands-on work, I learned how different neuroimaging outputs are represented and interpreted.

Three panels: structural MRI, functional network maps, and tractography colored by direction
The three kinds of output I learned to read.

Outcome

This understanding informed my later product and report-design decisions. It let me make choices about grouping, labeling and hierarchy based on how clinicians read the outputs.

02

Designing clinical brain-mapping reports

The challenge

Brain-mapping reports contained highly technical information: tract visualizations, network maps, statistical measures, radiologist interpretation and explanatory material. All of it needed to be presented coherently to clinical users.

Goal

Design report layouts that made complex neuroimaging information easier to navigate and interpret.

My approach

I designed clinical brain-mapping reports in Adobe InDesign. I established the information hierarchy and layouts for tract visualizations, network maps, z-scores, radiologist impression fields and contextual explainers.

The report layout with seven numbered design decisions
The report layout, recreated, with the reasoning behind its order and grouping.
Close-up of the tractography module: tract selector, two brain views, left and right measures, and the radiologist's impression field
A closer look at one module: tractography.
03

Defining product requirements

The challenge

The platform had a large number of interconnected workflows across different user roles. These included hospital administration, member management, studies, onboarding, patient uploads and neuroimaging outputs.

Goal

Create clear product requirements that translated the interface designs into expected behavior for engineering.

My approach

I defined user stories and acceptance criteria across 26+ screens. For each screen, I documented what it should display, how users should move through the flow, what actions should be available, and how the system should respond.

The requirements covered hospital administration, inviting and managing members, study setup, onboarding, DICOM upload, patient case management, and output selection for network maps and tractography.

26 screens across six flows, with one recreated screen and its six acceptance criteria
One screen with its acceptance criteria, recreated with placeholder content.

Outcome

I created a shared requirements reference that gave engineering clearer implementation expectations across a complex set of screens and user flows.

How it connects

The work followed one line: understand the neuroimaging domain, design the clinical reports, then define the product requirements. Knowing what the outputs meant shaped how the reports were organized, and designing the reports and screens made it possible to specify how the product should behave.