Jesse Jr Lim

Commodities Director · AI Infrastructure Builder · Founder
London, UK · decentre.io

Profile

Twenty years in physical commodities and LME futures trading — from trade administration through to Director — built a deep operational understanding of high-stakes, time-sensitive, consequence-heavy environments where traceability, precision, and speed of decision are not optional.

In parallel: a self-taught builder. A simple trade logging platform built in 2004 to automate daily contract processing was an early instinct that tooling could remove operational friction — one that has compounded into two serious software projects now in active development.

Currently building Pea, a governed agent runtime designed for enterprise environments that require auditable, memory-aware, and controllable AI infrastructure — and Decentre Studio, a generative creative suite with a patent-pending novel brush stroke synthesis engine. Both are self-funded, solo-built, and architecturally non-trivial.

The thread connecting twenty years of trading, two decades of self-directed learning, and two ambitious software builds is the same: a contrarian instinct for finding where existing systems are suboptimal and the persistence to build something better.

Core Competencies

Domain Expertise
Physical metals trading (non-ferrous scrap, secondary lead, zinc, tin ores, concentrates, ingot), LME futures, trade finance, incoterms, vessel logistics, annual contract management
Technical
Python, SQL, agent architecture, LLM systems, MCP protocol, vector databases (Qdrant), multi-LM consensus systems, generative image models (Stable Diffusion), rules-based trading systems
Leadership
Director-level P&L ownership, cross-functional team management, product management (Flutter mobile application), founder operating across product, engineering, strategy, and commercial simultaneously
Creative
Self-taught painter, commercial photographer and videographer — directly informing product design in Decentre Studio

Experience

Founder & Principal Engineer
2024 – Present
Decentre / Pea · London, UK

Building two concurrent software products from zero, solo, self-funded:

  • Pea — a governed agent runtime for enterprise environments. Architecturally layered system with a novel tiered memory model (L1/L2/L3 with delta-scored sedimentation), temporal binding across agent families, MCP-first capability dispatch, and full operator observability surfaces. Designed for high-accountability operational environments where auditability, traceability, and memory governance are requirements not afterthoughts. 200k+ lines of code, two weeks into second major version.
  • Decentre Studio — a generative creative suite combining a custom image editor, NLE video editor, and vector program. Includes a patent-pending generative brush stroke synthesis engine that replaces traditional bitmap stamping with real-time diffusion applied per stroke, with automated layer, mask, and compositing management. 600k+ lines of code across seven months.

Pending USPTO patent application covering the generative stroke synthesis process and stroke tokenisation method.

Current trading platform side project: rules-based signal system backed by a multi-LM consensus machine — not autonomous trading but signal generation with human execution.