Software Engineer · NLP & LLM Systems
AI Buddy Catalyst Labs, Ottawa · Previously Yugen Group, Birmingham

me = {
    "name"      : "Umme Abira Azmary",
    "role"      : "Software Engineer · NLP & LLM Systems",
    "currently" : "AI Buddy Catalyst Labs, Ottawa",
    "research"  : "Multilingual LLMs & knowledge conflict detection",
    "stack"     : ["Python", "HuggingFace", "PyTorch", "Next.js", "RAG"],
    "building"  : "bangla-text-toolkit · AI-native SaaS products",
   
}

Currently at AI Buddy Catalyst Labs (Ottawa), building AI-native SaaS and client products. Previously at Yugen Group (Birmingham) where I shipped a RAG system, n8n pipeline chatbot, and full authentication flows for their product Octobra!n.

I also do research — my published paper benchmarks how multilingual LLMs handle knowledge conflicts, evaluating BanglaT5, mT5, Qwen-2.5, DeepSeek-R1, and LLaMA-3.3 on parametric vs. contextual reliability.

NASA Space Apps Challenge 2022 Global hackathon	National AI Hackathon 2025 Top 8 of 102 teams
Academic Excellence BRAC University, Dhaka	Vice Chancellor's List 6 consecutive semesters

Production-grade Bangla NLP preprocessing in Python. Unicode normalization, tokenization, text cleaning, stopword filtering. Zero dependencies.

from bangla_text_toolkit import BanglaTextPipeline

pipe = BanglaTextPipeline.default()
pipe.run("আমি বাংলা ভালোবাসি!!! https://example.com")
# → 'আমি বাংলা ভালোবাসি'

Built because Python's \w doesn't match Bangla combining characters — most NLP tools silently corrupt Bangla text. This one doesn't.

NLP & ML

Engineering