Heya, I'm Junxian Ma, aka EdgeP - About Me Writings Projects GitHub CV

I am a CS student at Tsinghua University. My recent work centers on post-training for reasoning and alignment.

Research interests: language models, agents, reinforcement learning, reasoning, and alignment.

Here are some things I believe in:

• Reasoning models should be useful before they are impressive
• Post-training is where many systems reveal their real shape
• Agents need memory, taste, and restraint
• Good research taste compounds when you write things down
• Small tools can change the shape of a week
• Clear thinking is often a design problem

Here are some things I like:

• Language models, small agents, reward signals, clean abstractions, quiet libraries, fast feedback loops, command lines, elegant notes, diagrams that explain themselves, and people who ask precise questions.

Things I am working on:

• GitHub: code, notes, and unfinished ideas.
• CV: a more formal snapshot of my work.
• edgep.me: this small home on the internet.

Writings:

• Notes on learning draft
• What I am paying attention to soon

I love thoughtful messages and usually write back. Send me a message!