Portrait of Haojun Ma

Human-Centered AI · Interactive Systems

Haojun Ma

I am a Software Engineering senior at Zhejiang University. I research and build human-centered AI systems, with current work on efficient online multimodal recognition and interactive tools for accessible communication and creative practice.

  • CurrentZhejiang University
  • Past studyUC Berkeley · Fall 2025
  • DegreeB.S. Software Engineering · Jun 2027
  • Open toResearch + Internships

Selected Work

Research and prototypes, grounded in inspectable evidence

Sign2Text translation interface

Undergraduate Research → Independent Continuation · Sep 2025–Present

Sign2Text

An iOS interface and online continuous sign-language recognition research system balancing compute, latency, and recognition quality.

Contribution
Led a three-student team, built the SwiftUI workflow, and independently continued research on strictly causal, budget-aware scheduling.
Evidence
On Phoenix-2014T dev, a frozen adaptive baseline cut processed windows by 32.6% and repeated wall time by 29.5%, with a 0.19-point WER increase.
Online RecognitionMultimodal AIEfficient InferenceiOS
SceneZero product direction showing AR capture, structured creative input, and an inspiration library

Independent Continuation · May 2026–Present

SceneZero

An iOS spatial notebook for capturing multimodal inspiration and developing scene facts into reusable creative inputs.

Contribution
Independently repositioned and extended a course-originated prototype through multiple iterations.
Evidence
The current iOS target has a verified device build, passing unit tests, simulator launch, and a relaunch persistence smoke test.
Creative ToolsSpatial ComputingiOS
CHI social media research dashboard screenshot

Course–Lab Collaboration · May–Jun 2025

CHI Social Media Research Dashboard

A selective course–lab collaboration that turned 197 tagged CHI social-media papers into a linked-view dashboard.

Contribution
Built the Overview Panel and implemented chart-to-Sankey filtering within a three-person frontend team.
Evidence
Public source and deployed demo covering 197 tagged records; the project is archived.
Data VisualizationFrontend Engineering
Held-out test metrics for four sentiment-classification pipelines

Mentored ML Study · Feb–Mar 2024

NLP Model Comparison

A controlled comparison of classical sentiment pipelines on 4,000 IMDb movie reviews.

Contribution
Implemented preprocessing, model comparison, and validation-based tuning in a mentored study.
Evidence
TF-IDF + Logistic Regression reached 89.3% on a held-out 15% test split.
Sentiment AnalysisMachine LearningModel Evaluation

Explorations & Interests

Film, creative practice, and the experiments they lead to

Movie recommendation agent interface

Personal AI Experiment · Mar 2026–Present

Movie Recommendation Agent

Connecting film taste and Letterboxd context to an agent workflow with confirmation before account changes.

Poster for A Dream of Confucius

Collaborative Creative Exploration

AI Short Film

A collaborative exploration of generative tools, ritual, memory, and visual storytelling.

DaVinci Resolve editing workflow

Creative Practice

Video Editing

Small personal pieces for practicing pacing, framing, and visual storytelling.

Music · Team Sport

Personal Practices

Music & Team Sport

I return to instruments for the pleasure of learning sound, and to football for the shared rhythm of training and moving with a team.

Contact

Let’s build or investigate something useful

I am open to research collaborations and internship opportunities in human-centered AI and interactive systems.