Susie Fan

I built Story for Xixi as an AI reading buddy for my 20 months old daughter. Parents read a physical picture book with their child while Mimi, a playful cat character, joins the conversation. Mimi narrates each page, asks questions based on what the child can see, and builds on the child’s response. I wanted AI to make shared reading more engaging and personalized while keeping the physical book and parent child interaction at the center of the experience.

I started this project after reading research on early childhood development and reflecting on my own experience as a parent. The NIH describes the first three years of life as the most intensive period for acquiring speech and language skills, while the American Academy of Pediatrics recommends shared reading from infancy because it supports language, cognitive, social emotional, and literacy development. As a parent, I also found that creating a rich conversation around a book takes much more effort than simply reading the words on the page. I had to keep my toddler engaged, come up with age appropriate questions in the moment, interpret unpredictable responses, and introduce new ideas while still staying present in the story.

I first built a simple prototype that let Mimi add narration and questions to books, then tested it while reading with my daughter. During those sessions, I kept reaching for the laptop to repeat Mimi, tell Mimi that my daughter had responded, or move to the next interaction. Those actions interrupted the natural rhythm of reading, so I redesigned the experience to reduce device interaction. A parent can now hold up an open palm to ask Mimi to repeat herself, give a thumbs up when the child responds, or make a V gesture to continue. A parent can also simply turn the physical page, and Story for Xixi advances automatically. Mimi never tests or grades the child. She accepts speech, pointing, gestures, expressions, imitation, and other forms of engagement, then uses the parent’s signal to take the conversation one step deeper.

Technically, Story for Xixi uses GPT-5.6 Luna to analyze photographs of each physical book and extract printed text and visual page evidence. SQLite stores that evidence alongside child profiles, reading history, rewards, and feedback. A separate Authoring Agent powered by GPT-5.6 Terra uses an agent harness to assemble context, apply a versioned system prompt, validate Structured Outputs, and generate narration, questions, and follow ups. Long term memory comes from the child profile and reading history, including language ability, interests, and learning preferences, while short term memory combines the current book, page evidence, story continuity, and relevant child context. CosyVoice 3 pre generates Mimi’s Mandarin narration. MediaPipe Gesture Recognizer processes parent gestures locally in the browser, while deterministic image difference logic detects physical page turns. This architecture keeps generative AI in the content preparation and personalization layer rather than the live reading loop, which makes the experience faster and keeps the camera feed local.

I started Story for Xixi to make reading with my toddler easier, but I see a much larger opportunity over time. As a child grows, their AI reading buddy could remember what they understand, what interests them, and how they learn, then introduce increasingly sophisticated language, reasoning, numeracy, music, and other forms of learning. I want parents and teachers to remain at the center of a child’s development while AI gives more families access to the continuity and personalization of a devoted one on one tutor.

#AI #GenerativeAI #AIProduct #AgenticAI #EdTech #EarlyChildhood #ComputerVision #VoiceAI #ProductDesign #HumanComputerInteraction

An AI-powered platform that transforms project requirements and sketches into editable architectural floor plans and construction drawings. The image here is a sample floor plan generated by our prototype.

According to my interviews with 20+ architects, real estate developers, general contractors, and homeowners, creating construction documentation is repetitive, time-consuming, and expensive. Traditional workflows can take months, while professional design tools are often difficult for non-architects to use.

After prototyping and testing different approaches, I worked with my cofounders to develop ArchiteX AI to automate parts of the architectural planning process. The platform combines generative AI with algorithmic space planning to generate multiple floor plan options, optimize furniture layouts, and export editable DXF drawings, along with design rationale, cost estimates, and construction timelines.

Technically, ArchiteX AI combines a React-based interface with a Python generation pipeline. We used Claude through the Anthropic API to interpret spatial requirements, py3dbp to optimize furniture placement and prevent overlaps, and ezdxf to generate editable DXF floor plans that can be opened and refined in tools such as AutoCAD. The system connects AI reasoning with deterministic spatial algorithms to turn user requirements into both visual previews and production-ready vector drawings.

#AI #Architecture #GenerativeAI #ProductDesign
#Automation #FloorPlan #PropTech #ConstructionTech

How might we create an environment to help Michael, who's recently disabled, feel empowered about his current state of being so that he feels more confident to connect with others?

Our team has interviewed four disabled people and three staff from the Inglis House in Philadelphia and New York department of transportation and designed a new way for the disabled to play their favorite games.

Talk to the play station to register your customized gestures for left, right, forward, and backward. Play games with others and make new friends!

#GameDesign #InteractionDesign #UX #GestureRecognition

The first gen of our AI Campus Ambassador DINA.

According to interviews, most university students in China think JD as a traditional online retailer, not an innovative technology company. The company needs to change this perception to attract top candidates to join.

After prototyping and testing different online and offline solutions, we decided to focus on creating a memorable offline recruiting experience in the form of a "campus ambassador" that answers questions from students.

#AI #VoiceProduct #InteractionDesign #UI/UX #FacialRecognition #NLP #TTS #ASR

I started experimenting with VTubing (streaming with a virtual avatar) in 2021 as a way to make economics more visual, playful, and approachable. I created original 2D and 3D virtual avatars and used them to explain economic ideas through livestreams and videos, combining research, storytelling, character design, animation, and technology into one creative format. Here's a 3D character I built in Unity and controlled using Quest VR headset and hand trackers!

What began as a personal experiment grew into a community of 88K followers and more than 5M views!

I’ve been telling stories through songs since I was little, using music to capture people, places, and moments I want to remember. I have produced and published 8 original songs so far.

One of my favorite experiments is 《原梦纪》(Chronicle of Dreams), an original fan song I wrote, composed, arranged, produced in Logic Pro, and performed for Genshin Impact’s second anniversary. I sang alongside virtual singer Luo Tianyi (洛天依), bringing together my love of songwriting, music production, and virtual characters in one piece.

I like making everyday objects a little more delightful. Spife started with a simple observation: people rarely use the bottom-right corner of a spoon. I hand-carved a piece of mahogany into a spoon–knife hybrid, keeping the warmth and comfort of a wooden spoon while adding an edge that can cut, fillet, and pick out small fish bones. After six hours of sanding, it became one of those objects that was as satisfying to hold as it was to use.

Joy is light, playful, colorful, and carefree. Cooking is my way of expressing that simplest form of joy to my loved ones.