Engineer who builds
what shouldn't work, yet.

I'm Yichan, an engineering student at Imperial College London, currently #1 in my cohort. I work on computer vision pipelines, FPGA-accelerated systems, and the kind of low-level problems most people route around.

AI assistant launching soon. For now the buttons above are decorative.

#1
In cohort, Imperial EIE
68×
Peak FPGA speedup vs NumPy
90%
First-pass uplift, prod pipeline
01 / Experience

Things I've shipped, places I've broken.

2025 - Now

Software Engineer Intern, Computer Vision

SOTA AI · London

  • Rebuilt the core image-generation pipeline from scratch, compositing text labels independently of the diffusion model, landing a 90% lift in first-pass success and an 88% drop in turnaround time across production workloads.
  • Solved reliable text rendering across curved surfaces, rotated fonts, and complex packaging (a problem the wider industry had filed under "unsolvable"), opening up 30% more addressable market.
  • Deployed live for a real brand client across multiple SKUs and off-centre compositions. Hit zero manual correction steps across all tested edge cases.
  • Self-initiated a post-deployment security audit, caught a safety-override vulnerability, documented it in full and hardened the system before wider rollout.
Oct 2025 - Now

UTA, Digital Electronics & Computer Architecture

Imperial College London

  • Supporting 100+ students through labs and tutorials on memory systems, low-level architecture, and pipelined CPU design.
Jul - Sep 2025

Software Developer & Project Manager

Ideatec Software · Seoul

  • Led a 7-engineer team, owning backend APIs and the core AI orchestration model (Python, FastAPI, juFastMCP, IPaaS), automating 30+ workflows and cutting manual task time by ~40%.
  • Built a RAG retrieval system with embeddings for dynamic dataset access and context-aware responses.
  • Ran the agile cadence on AWS CodeCommit and presented results at the APEC SME & Venture Expo to international delegates.
Apr 2025

Macquarie Technology Insight Programme

Macquarie · London

  • Led a 4-person team to win Macquarie's internal technical building competition with an AI solution addressing urban overcrowding.
2023 - 2025

Satellite Operations & Administration

ROK Army · Icheon, South Korea

  • Restored encrypted satellite comms links that had been down for 18 months, ensuring 24/7 operational continuity.
  • Directed the upgrade of 300+ units to modern systems, boosting reliability and reducing downtime.
02 / Selected projects

Hardware, compilers, and other rabbit holes.

FPGA · HFT

Real-time Crypto Trading Engine

↗

Low-latency algo system on PYNQ-Z1 FPGAs. ~18× average / 68× peak speedup over optimised NumPy on live Binance order books. 57.6% live directional accuracy on 2-second forecasts.

SystemVerilog PYNQ-Z1 Python Binance API
FPGA · ML

Streaming Outer-Product Engine

↗

Rank-1 outer-product streaming, BRAM-resident, for live 13×13 matrix updates. 2× lower latency than BLAS-backed NumPy end-to-end across 8,000 streaming samples.

SystemVerilog BRAM Streaming
Compiler

C90 → RISC-V Compiler

↗

From-scratch C compiler targeting RISC-V assembly. Lex/Yacc parsing, AST construction, scoping analysis, codegen. Passes 100+ functional tests.

C++ Lex / Yacc RISC-V
CPU Design

RV32I Pipelined Processor

↗

5-stage RISC-V CPU with full hazard resolution, 2-way writeback cache, dynamic branch prediction (PHT + BTB), and single-cycle M-extension execution.

SystemVerilog RV32I + M Cache
03 / Toolkit

What I reach for, and how often.

Languages
Python / proficient C++ / working SystemVerilog / working Assembly / basic R
Systems & ML
FastAPI LangChain MCP RAG / Embeddings Diffusion pipelines FPGA / PYNQ
Tools
Linux Git AWS CodeCommit Jira Confluence Arduino
Off the keyboard
Korean / native English / native Data viz (ggplot2 / Shiny)
04 / Get in touch

Let's build
something good.

If you've got a problem at the edge of what's possible (hardware, vision, low-level), I'd love to hear about it.

Say hello
Email yk1122@ic.ac.uk GitHub github.com/YichanKim LinkedIn linkedin.com/in/yichankim