HomePortfolioPostsDashboard

Elijah Ng

elijah@0xeljh.com·github.com/0xEljh·0xeljh.com·Singapore

Summary

Machine Learning / Systems Engineer focused on training and inference optimisation. Experienced with PyTorch internals and kernel level work as well as upstream problem framing— turning business needs into production Vision/LLM systems. Proven ability to translate research concepts into measurable performance gains.

Skills

Training and Inference Optimisation: PyTorch internals, Triton, torch.compile, quantisation, CUDA (reading), FSDP2, HuggingFace ecosystem (accelerate, transformers, etc.), Unsloth, bitsandbytes, tinygrad

Machine Learning: PyTorch, finetuning, LLM/Vision pipelines, RAG, wandb, mlflow

Data & Pipelines: Prefect, Pandas, SQL, Pydantic, OpenTelemetry

Infrastructure: Docker, Nix, FastAPI, PostgreSQL, GCP, AWS/Cloudflare

Experience

Technical Cofounder — IBVC Inc. (Legal/Real Estate Tech)

2025–Present

  • Shipped production document understanding & LLM pipeline to extract structured data from unstructured records/filings, implementing schema validation, repair logic, and pipeline idempotence
  • Built a context-aware query-variant generator and results filter for skip-tracing. Improved contact hit-rate by 20x while capping cost to <$1/lead
  • Built Prefect flows aggregating 50+ heterogenous legal/market data sources into a unified data lake; processed 3k records/day with OpenTelemetry tracing and LLM-assisted error logging, driving 10x YoY growth in qualified leads
Early Software Engineer — Pacts (Crypto x Anti-Sybil)

2024

  • Owned app/frontend design and implementation; built on-chain analytics tooling the anti-Sybil airdrop platform
Technical Cofounder — MarinaChain (Crypto x Maritime Sustainability)

2022

  • Processed 1.3TB of raw telemetry data via Dusk to engineer features for a physics-informed maritime CO2 emissions model. Fused geospatial data with vessel specs to achieve explainable fuel and emissions estimates
Machine Learning Engineer — MindPointEye (Founded by inventor of ELMs)

2021

  • Improved YOLOX-tiny model mAP from 0.60 to 0.85 through data augmentation (SimCLR), LR scheduling, implementing a HPO pipeline, and optimizer implementation tweaks
  • Developed quantisation + compilation pipeline for YOLOX ONNX graphs to RKNN (Rockchip NPU)
  • Initiated and led regional team on semi-supervised image labeling and dataset curation via latent space analysis (using fine-tuned GAN embeddings), saving hundreds of team hours per project

Projects

Unsloth Challenge — MLPerf Puzzles

2025

  • Implemented custom Triton kernel for NF4 dequantisation, achieved 25% speedup over Unsloth baseline on T4
  • Enabled QLoRA fine-tuning with FSDP2 and torch.compile with no graph breaks
  • Implemented a memory-efficient backprop (inspired by cut-cross-entropy) that is compatible with GRPO
vamptutor.com — Vector-based card search for MTG

2025

  • Built a natural-language search for Magic: The Gathering cards using a fine-tuned qwen-embedding model
ETH Tokyo 2023 — Winner: Best Dashboard (3K USD)
  • Doubled down on analysis over visuals: shipped a functional Jupyter Notebook with aggregated analytics for 1inch Fusion resolver on-chain activity, execution profits, and gas spend
Dreambooth Optimization

2023

  • Reduced peak VRAM by 50% for Stable Diffusion fine-tuning through quantization and attention chunking. Turned client profitable by fitting training on 3080 instances from 3090s
Liquid Crypto Index Fund — Empire Group (HK fund)

2023

  • Curated a dataset of 2,000+ tokens from 2013–2023; developed backtests for systematic index fund strategies with simulated execution conditions (slippage; modelled from market data)

Open Source

  • Contributed bug fixes to PyTorch Lightning and bt (backtrader)
  • *Classic SGD: Reverted PyTorch SGD to original Sutskever formula for separable LR/momentum behavior

Education

National University of Singapore

2017–2021

BEng, Engineering Science. Minor in Computer Science.

Specializations: Computational Engineering, Biomedical Engineering

Honors with Distinction; A- median grade | 5 postgraduate modules

Final Year Project: Self-Organising Neural Networks

Internships: A*STAR (post-quantum crypto for ML) and DSO National Laboratories (opto-acoustic FEM solver)

Writing

  • Saving VRAM with Apple's Cut Cross Entropy — Triton kernel breakdown
  • Derivation: Cross-Entropy — First principles derivation
  • LR Scheduling and SGD — PyTorch SGD internals