An AI developer with a biology background, exploring the intersection of life, agents, and engineered systems.
"Boundaries are not endings, but beginnings."
At twenty, standing at the intersection of two worlds.
Two questions have always fascinated me:
- How life encodes itself
- How intelligence acts upon reality
Thus, my interests converge along three threads:
- AI for Biology: Using deep learning to understand biological information and sequences
- Embodied Intelligence: Moving agents from "reasoning" to "acting"
- Modular Systems: Building composable, scalable, and reusable engineering systems
I aim to build more than just code that "runs" — but systems that carry ideas, map reality, and continuously evolve.
- Biological sequence modeling (DNA / RNA / expression prediction)
- Agent systems and embodied intelligence
- Deep learning system design and modular abstraction
- Cross-disciplinary applications of AI and science
CNN-based DNA Expression Prediction
Using convolutional neural networks to predict expression patterns from DNA sequences, exploring the mapping between sequence structure and biological function.
Tech Stack: Python PyTorch CNN Bioinformatics
Highlights:
- Designed for DNA sequence modeling tasks
- Includes data preprocessing, model training, and evaluation pipeline
- Focuses on local pattern recognition in biological sequences
2. claw_arm
Agent-controlled Robotic Arm
.gif)
Integrating agent logic into robotic arm control, forming a closed loop between decision-making and physical execution.
Tech Stack: Python Agent Robotics Control
Highlights:
- Designed for embodied intelligence and intelligent control scenarios
- Focuses on "perception–decision–execution" system integration
- Explores the practical deployment of agents in real-world tasks
Building-block Deep Learning System
Decomposing and recomposing deep learning pipelines into modular components, reducing experiment costs and improving system reusability and scalability.
Tech Stack: Python Deep Learning System Design
Highlights:
- Emphasizes modular design of model training pipelines
- Designed for experiment management and rapid iteration
- Focuses on "how to build" rather than just "single-run results"
ATCG writes life's most ancient program, each base pair, an information encoding spanning billions of years.
0 and 1 carry the pulse of a new era, every loop, every gradient update, approaching the boundary of some "artificial understanding."
When the microscope's eyepiece meets the display's glow, I see:
- A river flowing for four billion years: from single cells to consciousness
- A river flowing for merely half a century: from computation to creation
The true miracle does not happen at the center of any single discipline, but in those boundary-blurring, collision-stirring, yet-to-be-named intersections.
Text is not merely a medium, but the projection of thought onto paper. Writing, for me, is not just expression — it is the process of organizing the world and reconstructing the self.
Theory that cannot be grounded cannot truly touch reality. I want to transform abstract ideas into systems that run, can be verified, and can be iterated.
Technology can carry us further, but what truly determines "why we move forward" remains internal judgment, values, and inquiry.
Kant said, two things fill the mind with ever-increasing awe —
The starry heavens above: The vastness and order of the universe, the motion and laws of all things
The moral law within: The depth and radiance of humanity, the freedom and dignity of the soul
Between exploring life and writing code, I often pause to gaze at the stars and examine my heart. Technology can reach the stars, but only reflection lets the soul take root.