Afifah Kashif

Undergraduate Capstone · 2025

Assistive Ointment Care Robot

Undergraduate group capstone project at the University of Washington, taught by Dr. Maya Cakmak. An assistive system on the Stretch RE2 mobile manipulator that lets people with limited mobility apply topical medication independently, with a custom brush attachment, programming-by-demonstration with ROS and a web-based GUI for area selection and emergency control. Designed around a human-in-the-loop model that preserves user agency in intimate healthcare contexts. Co-designed with people who manage chronic eczema, professional caregivers, and an occupational therapist.

Human-Robot Interaction · Co-Design Watch demo → Read paper →

Masters Dissertation · 2025-2026

Bias by Whose Definition?

MPhil in Human-Inspired AI at Cambridge, supervised by Dr. Jin Huang, Dr. Malak Sadek, and Dr. Hatice Gunes (AFAR Lab, HIVE Lab). Comparing dictionary-based and LLM-as-judge methods for evaluating social bias in conversational AI across three datasets and multiple levels of granularity, discovering that methods which share the label “bias evaluation” do not actually measure the same thing. Built a six-coder human ground truth benchmark, tested for self-evaluation effects when LLM judges score outputs from their own model family and surfaced a differential-refusal pattern of invisible allocative bias. Reframed bias evaluation as a question of measurement validity rather than method choice, providing recommendations for practitioners and regulators auditing deployed systems.

AI Evaluation · Fairness · Measurement Validity Paper TBA Code →

Human-AI Collaboration Research · 2026

Error Overlap in Human-AI Complementarity

Investigated why the most accurate model doesn’t always make the best human-AI team. Across three binary classification tasks, showed that error overlap between model and human (not standalone accuracy) predicts team performance, with the least accurate model sometimes forming the strongest team under learned deferral. Also examined how deferral strategies shift who receives human review, raising fairness concerns.

Algorithmic Decision-Making · Fairness Read paper → Code →

AI Research · 2026

AI x Neuroscience Investigation

Replicated and extended Goldstein et al.’s finding that the human brain and autoregressive language models share computational principles during speech comprehension. Using the Podcast ECoG dataset (9 subjects, 1,330 electrodes), reproduced evidence for next-word prediction, post-onset surprise, and contextual representation. A novel variance partitioning analysis showed that surprisal and contextual embeddings capture largely independent neural signals, supporting the view that prediction and representation are dissociable in the brain’s language system.

Computational Neuroscience · NLP · ECoG Read paper →

UX Research · 2024-2025

Wordplay, Coding for All

A multilingual, accessible, scaffolded coding environment built with Dr. Amy Ko at the UW Information School. Led template design (pre-structured code blocks students could drag and remix), coordinated an open-source community of nearly 80 contributors, and conducted classroom deployments in a partner school. Anchored the design in literature on inclusive edtech and educational governance.

Service Design · Ed-Tech · Accessibility See photo →

Community Work · 2025-present

IRC Digital Literacy Program

A program I founded with the International Rescue Committee and a local hospital to teach digital literacy to refugee communities in Seattle. I designed the curriculum, taught weekly classes, and secured grants for translation and access.

Curriculum · Accessibility · Community

Get in touch.

I am always down to chat about my work, interests, potential collaborations, or to grab a coffee :)