Ashwin Subramanian
Im a builder who is obsessed with design and the engineering behind the design.
CV
Where i sit
My research sits at the intersection of HCI and AI. On one side: how people understand, direct, and collaborate with AI systems, especially LLM-based agents, and whether the interface itself should be adaptive, generated by the agent on the fly to fit the task at hand, instead of forcing every interaction through the same fixed User Interface. On the other, a harder question underneath the design work: when an AI explains itself, how do we know the explanation is genuine, and not just persuasive?
Research Interests
Legible, adaptive human-AI collaboration.
How people direct and reason about LLM-based agents, and what it would take for those agents to generate their own task-specific interfaces on the fly rather than route every interaction through one static UI. A collaboration environment that bends to the task.
Genuine vs. persuasive AI explanation.
When a model explains its own reasoning, what separates an explanation that reflects the actual decision process from one that's just plausible-sounding. This sits closer to interpretability than to UX, but I think the two need to be solved together.
Research Work
Undergraduate thesis: predicting NFT market value from social and marketplace signals.
For my final year thesis, my team and I built a pipeline linking over 245,000 tweets to their OpenSea listings, then tested whether social signals like likes, engagement, and posting timing could predict an NFT's price better than the marketplace data alone.
Reproduced Filntisis et al.'s SPECTRE model for speech-aware 3D facial reconstruction.
Implemented the paper's lipread-aware loss and FLAME-based mesh pipeline end to end in to understand how visual speech signals shape reconstruction quality.
Background
I have an MS in Computer Science from UNC Charlotte (2025, GPA 3.7), focused on human-centered AI, machine learning, and visual analytics, and a BE in Computer Science from Anna University (2022, CGPA 8.2).
During my MS, I was a graduate teaching assistant for the department's Artificial Intelligence (ITCS 6150) Course under Dr. Dewan Tanvir Ahmed, running office hours and discussion sections on search, constraint satisfaction, and probabilistic models for 70+ students, and working with the instructor to refine assignments and rubrics.
On the industry side, I used to be a UX designer at Rebecca Everlene, working on MedKids, their K-12 anatomy education platform, alongside design consulting work for a handful of smaller startups. Before that, I co-founded a food-discovery marketplace startup as its founding product designer, taking it from a blank page to a closed beta over about a year and a half before the founding team scattered to different paths.
Earlier still, I spent three years as a programming mentor with my undergrad's ACM chapter, running weekly study groups and teaching outreach sessions in local schools.
My path here has moved through a few different angles on the same interest: an early software engineering internship gave me the infrastructure side, founding a startup, designing at Rebecca Everlene, and consulting gave me the product side, and my coursework and thesis gave me the research side. I'm trying to bring all three into one place.
What i have built
Designed and evaluated AI-assisted UX flows for a K-12 anatomy platform.
Ran usability testing to shape interaction flows and built a behavioral feedback pipeline that cut iteration cycles by 35%, reaching 1,000+ students.
Glide.
Designed and built an AI news app end to end, including the summarization model, personalization layer, and reading-format controls. Tested implicit swipe-based personalization against explicit controls with an HCI cohort, and found users wanted visibility over invisible automation.
Built an agentic research pipeline chaining browser automation, UI pattern search, and design tooling.
A structured competitive UX audit, covering a fixed set of dimensions (onboarding, core flows, navigation, monetization, and more) and synthesizing findings into comparison tables and cross-competitor pattern reports. Cut a 2 to 3 day research task to hours.
Co-founded and designed a two-sided marketplace app for real-time local discovery. (Truckerino)
Owning product design and frontend as the team's sole designer, and built data pipelines and dashboards for local vendors so they had a reason to trust the app before it existed. Validated across three rounds of research and a closed beta of 200+ users, choosing vendor-controlled location over automatic GPS to protect that trust.
Ashwin Subramanian
Im a builder who is obsessed with design and the engineering behind the design.
CV
Where i sit
My research sits at the intersection of HCI and AI. On one side: how people understand, direct, and collaborate with AI systems, especially LLM-based agents, and whether the interface itself should be adaptive, generated by the agent on the fly to fit the task at hand, instead of forcing every interaction through the same fixed User Interface. On the other, a harder question underneath the design work: when an AI explains itself, how do we know the explanation is genuine, and not just persuasive?
Research Interests
Legible, adaptive human-AI collaboration.
How people direct and reason about LLM-based agents, and what it would take for those agents to generate their own task-specific interfaces on the fly rather than route every interaction through one static UI. A collaboration environment that bends to the task.
Genuine vs. persuasive AI explanation.
When a model explains its own reasoning, what separates an explanation that reflects the actual decision process from one that's just plausible-sounding. This sits closer to interpretability than to UX, but I think the two need to be solved together.
Research Work
Undergraduate thesis: predicting NFT market value from social and marketplace signals.
For my final year thesis, my team and I built a pipeline linking over 245,000 tweets to their OpenSea listings, then tested whether social signals like likes, engagement, and posting timing could predict an NFT's price better than the marketplace data alone.
Reproduced Filntisis et al.'s SPECTRE model for speech-aware 3D facial reconstruction.
Implemented the paper's lipread-aware loss and FLAME-based mesh pipeline end to end in to understand how visual speech signals shape reconstruction quality.
Background
I have an MS in Computer Science from UNC Charlotte (2025, GPA 3.7), focused on human-centered AI, machine learning, and visual analytics, and a BE in Computer Science from Anna University (2022, CGPA 8.2).
During my MS, I was a graduate teaching assistant for the department's Artificial Intelligence (ITCS 6150) Course under Dr. Dewan Tanvir Ahmed, running office hours and discussion sections on search, constraint satisfaction, and probabilistic models for 70+ students, and working with the instructor to refine assignments and rubrics.
On the industry side, I used to be a UX designer at Rebecca Everlene, working on MedKids, their K-12 anatomy education platform, alongside design consulting work for a handful of smaller startups. Before that, I co-founded a food-discovery marketplace startup as its founding product designer, taking it from a blank page to a closed beta over about a year and a half before the founding team scattered to different paths.
Earlier still, I spent three years as a programming mentor with my undergrad's ACM chapter, running weekly study groups and teaching outreach sessions in local schools.
My path here has moved through a few different angles on the same interest: an early software engineering internship gave me the infrastructure side, founding a startup, designing at Rebecca Everlene, and consulting gave me the product side, and my coursework and thesis gave me the research side. I'm trying to bring all three into one place.
What i have built
Designed and evaluated AI-assisted UX flows for a K-12 anatomy platform.
Ran usability testing to shape interaction flows and built a behavioral feedback pipeline that cut iteration cycles by 35%, reaching 1,000+ students.
Glide.
Designed and built an AI news app end to end, including the summarization model, personalization layer, and reading-format controls. Tested implicit swipe-based personalization against explicit controls with an HCI cohort, and found users wanted visibility over invisible automation.
Built an agentic research pipeline chaining browser automation, UI pattern search, and design tooling.
A structured competitive UX audit, covering a fixed set of dimensions (onboarding, core flows, navigation, monetization, and more) and synthesizing findings into comparison tables and cross-competitor pattern reports. Cut a 2 to 3 day research task to hours.
Co-founded and designed a two-sided marketplace app for real-time local discovery. (Truckerino)
Owning product design and frontend as the team's sole designer, and built data pipelines and dashboards for local vendors so they had a reason to trust the app before it existed. Validated across three rounds of research and a closed beta of 200+ users, choosing vendor-controlled location over automatic GPS to protect that trust.
Ashwin Subramanian
Im a builder who is obsessed with design and the engineering behind the design.
CV
Where i sit
My research sits at the intersection of HCI and AI. On one side: how people understand, direct, and collaborate with AI systems, especially LLM-based agents, and whether the interface itself should be adaptive, generated by the agent on the fly to fit the task at hand, instead of forcing every interaction through the same fixed User Interface. On the other, a harder question underneath the design work: when an AI explains itself, how do we know the explanation is genuine, and not just persuasive?
Research Interests
Legible, adaptive human-AI collaboration.
How people direct and reason about LLM-based agents, and what it would take for those agents to generate their own task-specific interfaces on the fly rather than route every interaction through one static UI. A collaboration environment that bends to the task.
Genuine vs. persuasive AI explanation.
When a model explains its own reasoning, what separates an explanation that reflects the actual decision process from one that's just plausible-sounding. This sits closer to interpretability than to UX, but I think the two need to be solved together.
Research Work
Undergraduate thesis: predicting NFT market value from social and marketplace signals.
For my final year thesis, my team and I built a pipeline linking over 245,000 tweets to their OpenSea listings, then tested whether social signals like likes, engagement, and posting timing could predict an NFT's price better than the marketplace data alone.
Reproduced Filntisis et al.'s SPECTRE model for speech-aware 3D facial reconstruction.
Implemented the paper's lipread-aware loss and FLAME-based mesh pipeline end to end in to understand how visual speech signals shape reconstruction quality.
Background
I have an MS in Computer Science from UNC Charlotte (2025, GPA 3.7), focused on human-centered AI, machine learning, and visual analytics, and a BE in Computer Science from Anna University (2022, CGPA 8.2).
During my MS, I was a graduate teaching assistant for the department's Artificial Intelligence (ITCS 6150) Course under Dr. Dewan Tanvir Ahmed, running office hours and discussion sections on search, constraint satisfaction, and probabilistic models for 70+ students, and working with the instructor to refine assignments and rubrics.
On the industry side, I used to be a UX designer at Rebecca Everlene, working on MedKids, their K-12 anatomy education platform, alongside design consulting work for a handful of smaller startups. Before that, I co-founded a food-discovery marketplace startup as its founding product designer, taking it from a blank page to a closed beta over about a year and a half before the founding team scattered to different paths.
Earlier still, I spent three years as a programming mentor with my undergrad's ACM chapter, running weekly study groups and teaching outreach sessions in local schools.
My path here has moved through a few different angles on the same interest: an early software engineering internship gave me the infrastructure side, founding a startup, designing at Rebecca Everlene, and consulting gave me the product side, and my coursework and thesis gave me the research side. I'm trying to bring all three into one place.
What i have built
Designed and evaluated AI-assisted UX flows for a K-12 anatomy platform.
Ran usability testing to shape interaction flows and built a behavioral feedback pipeline that cut iteration cycles by 35%, reaching 1,000+ students.
Glide.
Designed and built an AI news app end to end, including the summarization model, personalization layer, and reading-format controls. Tested implicit swipe-based personalization against explicit controls with an HCI cohort, and found users wanted visibility over invisible automation.
Built an agentic research pipeline chaining browser automation, UI pattern search, and design tooling.
A structured competitive UX audit, covering a fixed set of dimensions (onboarding, core flows, navigation, monetization, and more) and synthesizing findings into comparison tables and cross-competitor pattern reports. Cut a 2 to 3 day research task to hours.
Co-founded and designed a two-sided marketplace app for real-time local discovery. (Truckerino)
Owning product design and frontend as the team's sole designer, and built data pipelines and dashboards for local vendors so they had a reason to trust the app before it existed. Validated across three rounds of research and a closed beta of 200+ users, choosing vendor-controlled location over automatic GPS to protect that trust.