$ whoami
AI & ML Engineer — MLOps, LLM pipelines, full-stack ML applications.
M.S. Computer Science (AI), Georgia Tech · CKA-certified · previously DevOps at Hyundai infotainment platform (Kubernetes across 20+ nodes).
Toronto, ON.
{
"name": "Dongjin Kim",
"focus": [
"ML / AI",
"MLOps",
"LLM Pipelines"
],
"location": "Toronto, ON",
"education": "M.S. CS · GT",
"certs": [
"CKA",
"TF Developer"
],
"open_to": true
}8 live applications · 2 coursework
End-to-end systems across investment operations, medical AI, agents, and reinforcement learning — every live project runs in production.
Live back-office platform — five operations modules, one market simulation.
5 ops modules · 1 live market sim
A live back-office platform — five operations modules over one continuously evolving market simulation: custodian reconciliation and break detection, NAV and performance reporting, multi-source data quality monitoring, trade-confirmation extraction with an SOP assistant, and trade-to-ledger double-entry accounting. Anchored to real FRED market data.
Chest X-ray triage, explainability and semantic search — one shared model.
14 conditions · 112k X-rays · 3 workflows on 1 model
A chest X-ray analysis platform over the NIH ChestX-ray14 corpus — DenseNet121 multi-label triage with FHIR patient context, Class Activation Map explainability showing which regions drove each prediction, and CLIP ViT-B/32 semantic image search. One shared model, three workflows.
A DQN agent and a Connect-4 game-AI arena in one workbench.
3 search algorithms · play live in browser
Two reinforcement-learning workbenches in one — a DQN agent learning Gymnasium control tasks (CartPole, Acrobot) with reward and loss streamed live over WebSocket, and a Connect-4 arena pitting alpha-beta minimax, Monte Carlo Tree Search, and an AlphaZero-style self-play network against you.
A tool-using agent that plans, acts, observes and self-corrects.
AST-sandboxed Python · self-corrects · streamed trace
A tool-using reasoning agent that runs a plan → act → observe loop — calling tools, reading results, and self-correcting until it solves the task. Tools include a safe calculator and a restricted Python sandbox (AST-validated, no file or network access). The full reasoning trace is streamed live.
Speech → transcription, sentiment, keywords — a full NLP pipeline.
Speech → 4 structured outputs in one pass
Upload audio → Whisper tiny transcription with timestamps, sentence-level sentiment via DistilBERT, keyword extraction, and speaking rate. Full NLP pipeline from raw speech to structured analysis report.
A gaze-tracked virtual desk lamp with object detection and memory.
6-DOF lamp · real-time gaze + object memory
6-DOF virtual desk lamp with real-time gaze tracking via MediaPipe iris landmarks, YOLOv8 object detection, and spatial memory queries answered by GPT-4o-mini grounded in a local visual store.
Regime-aware swing trading with multifactor ranking and paper execution.
Multifactor + HMM regime · 15-min paper-trade cycle
Regime-aware swing trading platform for US equities and ETFs. Multifactor ranking model (trend, momentum, relative strength, accumulation/volume) layered with HMM-based market regime detection, FinBERT sentiment analysis, and FRED macro inputs. Paper trading engine with entry sizing, stop/TP/trailing-stop, and a 15-min scheduled cycle.
A grounded RAG chatbot for the Korean Consulate in Toronto.
216 official posts · BM25 + embeddings hybrid
RAG-based civil service chatbot for the Consulate General of Korea in Toronto. Hybrid BM25 + OpenAI embedding search over 216 official bulletin posts (passport, visa, notarization, military service, and more), answered by GPT-4o grounded strictly in official content — with source links and disclaimer.
CNNs, RNNs and attention implemented from scratch in PyTorch.
Implemented CNNs, RNNs, and attention mechanisms from scratch in PyTorch for image classification, sequence modelling, and transfer learning. Reproduced foundational architectures and benchmarked against pretrained baselines.
A market simulation framework with Q-learning strategy learners.
Built a full market simulation framework implementing Q-learning and random forest strategy learners. Evaluated portfolio performance using Sharpe ratio, cumulative return, and drawdown against buy-and-hold baselines.
Dec 2025 — Present
Toronto, ON
Consulate General of the Republic of Korea
Political and Economic Affairs
Feb 2022 — Dec 2022
Seongnam, South Korea
Luxoft
Hyundai Motors — Infotainment Platform
Mar 2021 — Jul 2021
Suwon, South Korea
CSI Vision
May 2026
M.S. in Computer Science — Artificial Intelligence specialization
2022
B.S. in Computer Science
2017
B.B.A. in Business Administration
$ status --open-to-work