穆再排尔·穆合塔尔

03 About

Nanjing / 2026

A practice of looking closely.

I am Muzapar Muhtar, a mechanical engineering undergraduate at Nanjing University of Aeronautics and Astronautics. I work on computer vision for autonomous systems — the unglamorous middle layer where a model, a camera, and a physical machine have to agree.

Nanjing, China
32.0603° N / 118.7969° E
mzpr00@163.com

A black-and-white photograph of Muzapar Muhtar setting up a camera and laptop on a tripod outdoors.
PL. 05Plate — self, at work · Nanjing · fieldwork

My research examines selective cross-modal alignment in vision-language models. In a controlled set of 6,000 physical videos, I found that spatial position is linearly decodable from Qwen2.5-VL’s representations while velocity remains entangled in a manifold the model does not spontaneously use. The finding is narrow, and that is the point: it is easier to trust a claim when you know exactly where it stops.

My engineering work lives on the edge. I built a three-tier landing pipeline for drones, led the label audit that found 367 flipped boxes in a 1,999-frame dataset, and shipped a TensorRT / OpenVINO path to Jetson. I also built Nuclear Boy, an Android AI coding agent with a Python sandbox and twenty hardware controls, because I wanted to know what an agent feels like when it lives where you do.

I write these notes to keep the gap between a metric and a meaning visible. The site is also a small design exercise: a catalogue, a grid, a few typefaces, and as little machinery as possible.

Public record / 2022—2024

Four rooms · one note

Before the model

Four rooms between 2022 and 2024: a study session, a debate, a student-union defence, a summer classroom. Before the model, I learned to speak in rooms.

Read the record
  1. Opening of the 20th Party Congress The first room
  2. “Engineering × Humanities × Foreign Languages × Mathematics” four-school debate · preliminary The argument room
  3. Outstanding officer defence · School of Mechanical and Electrical Engineering Student Union, 2023—2024 The working room
  4. Red Heart Knot · summer public classroom (fifth year) The teaching room

01 Now

Currently

  • 01Preparing the NeurIPS 2026 workshop poster.
  • 02Fine-tuning Qwen-8B with QLoRA on a personal corpus.
  • 03Writing field notes on evaluation and error.

02 Practice

What I do

  • 01Computer vision for autonomous systems
  • 02Vision-language model interpretability
  • 03Edge deployment: TensorRT, OpenVINO, Jetson
  • 04Human-machine interfaces for small screens

03 Education

Education

2023.09 — 2028.06 (expected)

Nanjing University of Aeronautics and Astronautics

B.Eng. in Mechanical Engineering

Coursework: automatic control, C++, robotics, machine vision. Third-class scholarship; outstanding student cadre; class monitor.

04 Research

Selected research

05 Timeline

A working chronology

NeurIPS 2026 Workshop

Extended abstract accepted to NeurReps; poster presentation.

Autonomous landing

Project completed and rated excellent; detector, label audit, and edge deployment open-sourced.

Jittor Challenge

Temporal link prediction: 0.9357 → 1.2443; Val-Hot benchmark released.

Nuclear Boy

Android AI coding agent; first public release.

Jiangsu Robotics Competition

Vision-guided AGV, 11th of 23 teams in the off-road track.

NUAA

Began B.Eng. in Mechanical Engineering.

06 Tools

Tools

Languages
Python · C++ · Kotlin
Vision & learning
PyTorch · Jittor · YOLO · OpenCV
Deployment
TensorRT · OpenVINO · Jetson · STM32
Mobile
Jetpack Compose · Hilt · Room
LLM
GPT from scratch · QLoRA · DeepSeek API · SSE
Design & CAD
SolidWorks · typography · grid systems

07 Languages

Languages

Mandarin
Level 2A
Uyghur
Native
English
CET-6

08 Honors

Honors & service

  • 01NeurIPS 2026 Workshop (NeurReps) — accepted, poster
  • 02National Bronze, Nippon Paint “Color for Love” teaching award — team leader, 15 people
  • 03University third-class scholarship; outstanding student cadre
  • 04Jiangsu Robotics Competition — 11th of 23, off-road track
  • 05University innovation project — rated excellent