WriteRight began as an academic proposal co-authored with WenXi Wu (PhD student, Special Education) for a Spring 2026 course project. The evaluation plan below was proposed but not conducted. The interactive build linked here is a solo portfolio prototype with simulated handwriting analysis — not real computer vision.
An AI-powered writing intervention concept for adults with dysgraphia — built to grow real skill, not just confidence.
WriteRight is a cross-disciplinary final project pairing HCI design with special education research — an AI-powered writing intervention concept for adults with dysgraphia. The core question driving it: how do you build a tool that grows real handwriting skill over time, without inflating confidence the skill hasn’t earned?
PhD student in Special Education whose prior work studies video modeling and AI-generated video for autistic learners. Brought a qualitative and mixed-methods research design background and led overall direction and oversight.
Contributed intervention design and data analysis — informed by prior work managing PHNX, an mHealth intervention app for minority youth, built with UMD’s CREATE Lab.
Adults with dysgraphia face a persistent skill-development gap: most handwriting support is either clinical — occupational therapy, often infrequent and appointment-bound — or absent entirely once school-age accommodations end. Existing digital tools tend to evaluate handwriting rather than build it, giving users a score with no clear path to improvement, and no way to track whether confidence and actual skill are moving together.
Tools that boost a user’s confidence without producing real skill gains can do active harm — making someone feel improved while their underlying handwriting ability stays flat.
How might we help adults with dysgraphia build real handwriting skill over time, without inflating confidence the skill hasn’t earned?
Every design decision was required to connect to peer-reviewed research rather than intuition. Three readings shaped the core design logic — each surfacing a specific risk the design then had to answer for.
Confidence gains from AI interaction can be domain-dependent and, critically, unwarranted — a user can feel more capable without being more capable. This became the central risk WriteRight was designed to avoid.
Informed how AI-assisted creative and skill tools should be framed to avoid undermining user agency.
Structured AI representations function as guardrails that focus users without removing their autonomy — directly shaping the decision to present feedback within defined categories rather than as open-ended AI output.
Each interaction principle maps directly back to a finding from the research — the design is the argument for how to avoid the false-confidence trap.
Feedback and exercises are presented within defined categories and stages rather than freeform AI output, so users always know what they’re working on and why — responding to Ding et al.’s guardrails finding and the documented cognitive-overload barrier for adults with dysgraphia.
The AI persona is a writing coach, not a grader. Feedback is always framed as what to do next, never as what went wrong — a direct response to Reich & Teeny’s finding that AI can act as a high-credibility social referent in rule-governed skill domains.
Every piece of encouraging feedback is paired with the underlying objective metric that supports it — preventing the false-reassurance risk Reich & Teeny identified while still preserving motivational framing.
Users photograph a handwriting sample; the system analyzes letter formation, spacing, line alignment, and legibility, with color-coded (red/amber/green) feedback and plain-language explanations.
Six-week adaptive plans that shift emphasis based on a user’s evolving strengths and weaknesses across scans.
Short, targeted practice sessions (10–15 minutes) designed for sustainable daily engagement.
A longitudinal dashboard, with the option to share data directly with an occupational therapist or writing specialist — positioning the app as a supplement to clinical care, not a replacement.
This evaluation plan was proposed as part of the academic project and was not conducted — no usability sessions, skill assessments, or specialist interviews were run.
Think-aloud usability study (6–8 adults with dysgraphia) — would test whether the interaction design functions across core tasks: scan upload, reading feedback, navigating the action plan.
Pre/post skill assessment — standardized writing samples at baseline and after six weeks, scored blind by trained raters using the DASH instrument, to measure objective skill change rather than self-report.
Self-efficacy measurement at baseline, midpoint, and post — would track whether confidence and objective skill move together, directly testing the coach-framing and objective-anchoring decisions.
Structured specialist interviews — would validate the clinical soundness of the skill categories, exercise progressions, and specialist-sharing feature with occupational therapists and writing specialists.
To make the design rationale tangible beyond the written proposal, I independently rebuilt WriteRight as a standalone web prototype — a four-tab Next.js application (Scan, Progress, Tips, Profile) carrying the same three interaction principles into a working UI.
A conceptual portfolio piece, built solo and separately from the class project. The handwriting analysis is simulated for demonstration — it is not real computer vision, and the proposal’s AI Photo Scan Analysis has not been built with real CV.
Theory-grounded design holds up under scrutiny. Tying every interaction decision to a specific research finding made the rationale defensible when questioned — a habit worth carrying into professional UX work.
Cross-disciplinary partnership sharpens scope. Working alongside a special education researcher meant balancing HCI interaction-design instincts against rigorous qualitative and mixed-methods research standards.
A working prototype is not the same as a validated one. Building the portfolio MVP made the concept tangible, but the case study has to be honest that “AI-powered” describes the proposed design, not a built or tested system.