g/ A BUILDER'S TIMELINE @cryptoSUN2049

Experience & direction

  1. EARLIER01 / SYSTEMS

    Cloud systems engineering

    Tencent Cloud · OS team

    Worked on cloud operating systems and infrastructure. The systems layer came before the model layer.

    • Cloud infrastructure
    • Systems engineering
  2. FROM 202302 / WEIGHTS

    LLM training & post-training

    xDAN AI · Founder & model training engineer

    Led our team's work on the xDAN model family: training, alignment, and the engineering around model releases.

    • SFT / DPO
    • MoE
    • Reward modeling

    Selected team releases ↴

    2023xDAN-L1-Chat-RL-v17B chat model · SFT + DPO 2024xDAN-L3-MoE141B-parameter MoE release 2025xDAN-L2 · Qwen2.5-32B RM32B release in the xDAN model collection
  3. NOW CURRENT FOCUS03 / AGENTS

    Building Agent Native

    xDAN AI · Founder

    Building products and systems around agents: tools, execution, feedback, and real-world tasks. Turning model capabilities into useful work.

    • Agent infrastructure
    • Tools & execution
    • Evaluation

    Current work includes OpenFinClaw, a financial agent project.

    xDAN-DSH-uni-agentWork on a uni-agent fork
  4. NEXT04 / LEARNING LOOPS

    Self-improving models

    Research direction · metaRSI

    I want to close the loop: agents that don't just execute, but turn experience into better models.

    How can agent experience become measurable gains in model capability?

    An exploration direction. The next thing to build.