Skip to content
View gazelle93's full-sized avatar
13 followers · 20 following

Achievements

Achievement: Starstruck

Achievements

Achievement: Starstruck

Block or report gazelle93

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
gazelle93/README.md

Hi, I'm Mingyou 👋

I build AI systems from the problem statement to production and beyond.

Getting the model to work is rarely the hard part. Defining the right problem is harder, and keeping the system trustworthy once real users depend on it is harder still.

Currently an AI engineer at Potloc in Montreal. Before that, Lead NLP Engineer at Valital, running the NLP roadmap for AML and KYC and setting up the evaluation and monitoring practice the team ran on.

Things I do in public

  • decision-models-under-pressure An independent benchmark of seven decision models under three production stresses: growing candidate lists, shuffled option order, and harder distractors. Pre-registered before the first API call, with the dataset and per-call ledger published. Shuffling the answer options alone changed 14.6% of one model's decisions, and a fixed order does not fix it. The study also reversed the conclusion I started with, and the repo keeps the original reasoning rather than quietly rewriting it.

  • Assessing LLMs' Ability to Navigate Cultural Knowledge Conflicts (C3NLP 2024, non-archival). QARV, a 671-question benchmark for how LLMs handle conflicts between U.S. and Korean perspectives.

  • CLaC at SemEval-2020 Task 5 Multi-task stacked Bi-LSTMs for detecting the span of antecedents and consequents in counterfactual statements.

Education

  • MSc Computer Science, Concordia University. Thesis on input representations and classifiers for relation extraction, across SemEval-2010 Task 8, TACRED, Re-TACRED, and BioCreative VII (DrugProt).
  • BEng Computer Engineering, Hongik University

English and Korean. Reachable on LinkedIn

Popular repositories Loading

  1. Transformer-Various-Positional-Encoding Transformer-Various-Positional-Encoding Public

    This project aims to implement the Transformer Encoder blocks using various Positional Encoding methods.

    Python 24 2

  2. Multiclass-Focal-loss-pytorch Multiclass-Focal-loss-pytorch Public

    This is an implementation of multi-class focal loss in PyTorch.

    Python 10 1

  3. llm-fine-tuning-sft-lora-qlora llm-fine-tuning-sft-lora-qlora Public

    Practical examples for fine-tuning large language models (LLMs) with SFT, LoRA, and QLoRA using Hugging Face Transformers and PEFT.

    Python 8 1

  4. Attention-Various-Positional-Encoding Attention-Various-Positional-Encoding Public

    This project aims to implement the Scaled-Dot-Product Attention layer and the Multi-Head Attention layer using various Positional Encoding methods.

    Python 5

  5. decision-models-under-pressure decision-models-under-pressure Public

    Seven decision models, measured as the candidate list grows, the option order changes, and the wrong answers stop being obvious.

    Python 4

  6. charCNN charCNN Public

    This project aims to implement the charCNN word embedding method that is leveraged in ELMo and characterBERT.

    Python 3