Ayse Demir Portfolio

Born and raised in Turkey, Ayşe Demir is a San Francisco–based data science lead, computational artist and yoga teacher.

For nine years, she led large-scale causal inference, applied machine learning, data visualization and ML explainability projects at Amazon Web Services, Gap and Old Navy E-commerce. She has taught data visualization and creative coding at organizations including Gray Area Foundation for the Arts, Salesforce/Dreamforce, and Navajo Technical University’s Risk & Resilience Lab.

Her current work focuses on making complex systems interpretable through experimentation and visualization.

Selected works from teaching and speaking.

Course & Workshop Designs

Visualizing Memory-State Transitions in Transformer Models, Protocol Symposium (Protocol Institute, Upcoming 2026)

Operationalizing and interpreting phenomenological time in transformers through model dynamics.

Visual methods for studying attention, KV-cache retention and memory-state transitions.
Utilizing change-point detection as a method of identifying retained context shifts during inference.

Artificial Intelligence & Creativity, (DDI Akademi, March 2024)

A four-week course on machine agency and algorithmic creation, structured around the Leibnizian–Cartesian distinction between learning, combinatorial and rule-based models of mechanistic science.

Week 1: History of Machine Agency

Examining automata, cybernetics, and early AI, this week explored the cultural narratives that have shaped our perception of “machine life” and the creator’s role.
Week 2: The Evolution of Digital Culture

Tracing the past fifty years of computing and internet history, algorithms and machine learning to understand how we arrived at our current digital landscape.
Week 3: Creating with Algorithms

Exploring what it means to create with algorithms; algorithmic aesthetics, the aesthetics of indeterminacy, outliers, and noise; and how these artistic practices bring up notions of free will.
Week 4: Creating with Agents

Exploring what it meant to co-create with supervised and unsupervised learning models. Practicing prompting as a form of curation and expression, and concluding with a final discussion on creative control and agency.

Dreamforce Speaker, Visualizing Climate Data (Salesforce, September 2022)

Speaker at Salesforce Dreamforce 2022. Presented advanced data-storytelling techniques in Tableau, transforming complex climate data into clear, actionable narratives.
Prototyped novel views and self-joined subsets to get from a single dataset.
Showed how accessible, well-designed visualization can make complex data legible to non-technical audiences.

Creative Coding Intensive (Gray Area Foundation for the Arts, September 2020 - October 2023)

This intensive course focused on creating interactive environments. Students learned to capture physical data such as motion, touch, pressure, proximity and audio levels and to convert them into interactions and custom visuals.

Artist Talks (MUTEK, Google Art Week, 2020 - 2021)

Artist talks over years have centered on two key areas:

Using software to combine abstract art and data visualization.
Designing narratives that balance linear clarity (structured, guided) with nonlinear exploration (open, multi-path).

These works have been featured at New Art City, MUTEK 2020 and Google Art Week 2021.

Data Visualization Design (Gray Area Foundation for the Arts, September 2020)

This course traced the evolution of data visualization, from early statistical graphics to modern, web-based systems. Interaction was treated as an encoding, where controls, views and feedback loops became an integral part of how narration is made. The curriculum divided into two parts:

Part I (Theory): history, ethics, information architecture and visual encodings.
Part II (Practice): encoding of data into running systems with Python, Plotly, Dash.

Visual experiments explore abstraction and form through computation

Traditional software often produces abstract patterns detached from physicality. My explorations ask the reverse: can computation embody softness, intuition or ambiguity? I work across interpretability plots, p5.js, TouchDesigner, Blender and prompting to test when algorithmic output crosses from pattern into presence. My goal is to move beyond cold abstraction toward visuals that feel spatially grounded. Because each tool has its own aesthetic character, the final form is always a negotiation between intention and the tool’s inherent nature. Engaging with algorithms this way sharpens my intuition about the boundary between what a tool generates and who it interacts with.

Alongside my visual experiments, I keep a small curation page: a rotating set of images I posted on and off for the past three years. It mixes my own code-based and architectural forms with Deleuzian diagrams, historical models and other references. The selections are a way of thinking through patterns across media, showing how a sketch, a diagram or a fragment of code can inhabit form and perception.

I also explore prompting and AI tools as processes of curation rather than creation from scratch. I look for outputs that evoke a feeling I can recognize, sensing the moment an image shifts from diffusion noise into something with presence.