Skip to content
YJYeonwoo Jeong

Available for conversations

Optimization EngineerProblem structuring, system architecture, performance

I take vague, tangled problems, give them a structure worth solving, design the system around that structure, and then make it fast. Nine years doing this across medical AI, trading, robotics, and developer infrastructure.

EmailGitHubnpmLinkedInSee the work
Yeonwoo Jeong

Yeonwoo Jeong · Seoul, South Korea

7
Publications

ICML · CVPR · AISTATS

9
Years shipping

research → production

4
Open-source tools

agent infrastructure

SNU
Ph.D.

Combinatorial optimization

01 — About

How I work

  1. 01

    Structuring the problem

    Hard requirements arrive as a wish, not a specification. Deciding what is actually being optimized, and under what constraints, is where most of the leverage is — a problem framed well is half-solved, and framed badly no amount of engineering rescues it.

  2. 02

    System architecture

    Once the shape of the problem is clear, the system should follow it rather than fight it: where state lives, what runs on device versus server, which boundaries stay rigid so everything behind them can keep changing.

  3. 03

    Optimization

    Then make it fast — training that finishes in hours instead of years, a search space pruned until routing lands in sub-100ms, parsing cut by 10×, events handled at microsecond scale. A Ph.D. in combinatorial optimization is what makes this and the first step the same skill.

Selected results

  1. years → hours

    Reformulated optimal decision trees so industry-scale instances became trainable · DeepMetrics

  2. sub-100ms

    Near-optimal routing across hundreds of smart contracts via graph search with pruning · Eisen Finance

  3. 10×

    DOM parsing speedup in a hybrid LLM + structural web extraction system · Wordbricks

  4. microseconds

    Event handling for high-frequency execution on Polymarket · Alphanonce

  5. real-time

    On-device vs. server workload placement across a humanoid robot compute stack · Ludo Robotics

02 — Selected work

Agent infrastructure

All projects

  1. negotium

    TypeScript · 2026 — Present

    Host-agnostic multi-agent runtime

    maestrojeong/negotium1npm i negotiumv0.2.27
    • Turns a single machine into a durable multi-agent node, running Claude Code, Codex, and Maestro backends with provider-native sessions preserved across turns.
    • Generalizes Clawgram's orchestration into a reusable core — composable modules and local-first encrypted state behind a single adapter boundary, so new hosts plug into the same runtime.
    TypeScriptBunSQLiteMCP

  2. maestro-agent-sdk

    TypeScript · 2026 — Present

    Embeddable TypeScript agent SDK

    maestrojeong/maestro-agent-sdk9npm i maestro-agent-sdkv0.2.3
    • A composable agent SDK that reaches Claude Code-level agentic coding performance on non-Anthropic backends (DeepSeek, Kimi) — evidence that the agent loop, not the model, is the differentiator.
    • Sessions, memory compaction, MCP tools, guardrails, and subagent delegation are independently composable behind one streaming interface.
    TypeScriptESMMCPBun/Node

  3. clawgram

    TypeScript · 2025 — Present

    Telegram control layer for Claude Code

    maestrojeong/clawgramnpm i clawgramv2.7.1
    • In daily use by practicing lawyers on real case work. Private Mode masks PII with a pre-LLM hook before any prompt reaches the model, then unmasks on the response — confidential documents never leave the device unredacted.
    • A multi-tenant orchestration layer maps Telegram forum topics to persistent, isolated Claude Code sessions, with MCP primitives that let sessions exchange context and delegate asynchronously.
    TypeScriptBunFastAPIMCPPlaywright

03 — Experience

Where I've built

Full history

  1. Ludo Roboticsnow

    Apr 2026 — Present

    Physical AI Engineer · Seoul, KR

    • Building the AI agent harness for the humanoid robot brain — orchestrating tool-use across perception, planning, and control, and serving trained VLM policies in the loop.
    • Optimized CPU/GPU allocation and latency across the robot's compute stack, deciding workload placement (on-device vs. server) under real-time constraints.
    PythonPyTorchVLM/VLA

  2. Alphanonce

    Sep 2025 — Apr 2026

    Software Engineer · Seoul, KR

    • Developed and operated high-frequency trading strategies on Polymarket — real-time ingestion, signal pipeline, and low-latency execution with microsecond-scale event handling.
    • Built an AI agent that analyzes on-chain wallet transactions to surface alpha signals directly from public blockchain state.
    FoundrySolidityTypeScriptPython

  3. Wordbricks

    Apr 2024 — Jul 2025

    Deep Learning Engineer · Remote · Palo Alto, CA

    • Designed a hybrid web extraction system pairing LLM content reasoning with deterministic DOM parsing — page structure constrains and verifies the model's output, with DOM optimization delivering a 10× speedup.
    • Designed web macro automation that generalizes one user recording into many varied replays, so a single demonstration covers a whole class of pages.
    TypeScriptNext.jsNestJSAWSCloudflare

04 — Research

Publications

All papers
  1. NeurIPS 2025

    NEXT-EVAL: Next Evaluation of Traditional and LLM Web Data Record Extraction

    S. Kim, N. Kim, Y. Jeong

    BibTeX
  2. ICBC 2023first author

    Efficient Liquidity Providing via Margin Liquidity

    Y. Jeong, C. Jeoung, H. Jeong, S. Han, J. Kim

    BibTeX
  3. ICML 2023

    Efficient Latency-aware CNN Depth Compression via Two-stage Dynamic Programming

    J. Kim, Y. Jeong, D. Lee, H. O. Song

    BibTeX
  4. AISTATS 2022first author

    Optimal Channel Selection with Discrete QCQP

    Y. Jeong, D. Lee, G. An, C. Son, H. O. Song

    BibTeX
  5. ICML 2019first author

    Learning Discrete and Continuous Factors of Data via Alternating Disentanglement

    Y. Jeong, H. O. Song

    BibTeX
  6. CVPR 2019first author

    End-to-end Efficient Representation Learning via Cascading Combinatorial Optimization

    Y. Jeong, Y. Kim, H. O. Song

    BibTeX
  7. ICML 2018first author

    Efficient End-to-end Learning for Quantizable Representations

    Y. Jeong, H. O. Song

    BibTeX