class LaibaMazhar:
role = "Senior ML Automation Engineer @ AiPixal (Lahore, PK · on-site)"
also_at = "Data Developer & Analyst @ CloudWorks (Texas, USA · remote, part-time)"
also = "AI / ML Engineer · Data Scientist · Researcher"
education = "BS Data Science, FAST-NUCES ('22 → '26)"
based_in = "Lahore, Pakistan 🌍"
focus = [
"LLM agents · RAG · prompt engineering",
"Adversarial ML & federated-learning security",
"Explainable AI (SHAP) · intrusion detection",
"Batch + streaming data platforms at scale",
]
stack = {
"ml": ["PyTorch", "TensorFlow", "scikit-learn", "XGBoost"],
"llm": ["LangChain", "Hugging Face", "OpenAI", "Groq"],
"data": ["Spark", "Kafka", "Airflow", "Databricks", "dbt"],
"eng": ["Python", "SQL", "TypeScript", "FastAPI", "Docker"],
}
def currently(self) -> str:
return "Building ML automation at AiPixal & data products at CloudWorks 🚀"- 💼 Currently Senior Machine Learning Automation Engineer at AiPixal (Lahore, Pakistan — on-site, full-time) since August 2026
- 🌐 Also part-time Data Developer & Analyst at CloudWorks (Texas, USA — remote), building data pipelines, models and analytics
- 🔬 Researching adversarial robustness in federated learning, explainable IDS, and hallucination mitigation in multi-agent LLM systems
- 🤖 Building production RAG pipelines, autonomous agents, and end-to-end ML services
- 🏗️ Engineering batch & real-time data platforms with Spark, Kafka and Airflow
- 🎨 Founder & artist at Rangrayze, an art gallery for Sufi-inspired calligraphy and canvas work
- 🌱 Founder of airRTH, a community climate initiative running sustainability campaigns and environmental programs
- 🏆 Winner — Big Data Quest Hackathon, DataFest / NaSCon
- 🎓 Stanford University Machine Learning (Coursera, Andrew Ng)
- 🗣️ English (fluent) · Urdu (native) · German & Arabic (basic)
- 🌐 Full portfolio at laiba-mazhar.vercel.app
- 📫 Reach me at laibamazhar.000@gmail.com
🧠 AI · ML · LLMs
⚙️ Data Engineering · Big Data
💻 Languages · Backend · Cloud
📊 Analytics & BI
| 📄 Work | Focus | Headline Result |
|---|---|---|
| ICS — Adversarial Poisoning Defense in Federated Learning | Independent Class-wise Coherence Score against label-flipping & backdoor attacks, using MI-FGSM adversarial probing | 75.02% accuracy at 19.75% ASR on MNIST — beats global-coherence baselines by 5–7% accuracy with 30–40% lower ASR across 15 clients |
| Explainable Multiclass Intrusion Detection | XGBoost + TreeSHAP with a severity-scoring engine for SOC deployment (NSL-KDD, UNSW-NB15) | 99.90% avg accuracy, +12.28% macro-F1 over a Random Forest baseline, 2.44× stability gain over 5 seeds |
| LLM-Driven Autonomous Trading Agents | Analyst → Critic → Decision pipeline on Llama-3.3-70B (Groq) for hallucination mitigation in financial reasoning | 80% recommendation-adjustment rate and HDR 1.0 across AAPL, MSFT, TSLA, GOOGL, AMZN |
A multi-agent system that runs a micro-startup end to end — idea → GitHub PR → Slack launch post → cold outreach email, with no human in the loop. |
Full-stack school platform: React + Vite + TypeScript on Supabase (Postgres/Auth/RLS) — students, staff, timetable, attendance, fees, exams and dashboards. |
Robust, explainable multiclass intrusion detection with TreeSHAP, multi-seed evaluation and severity-aware alert scoring built for real SOC workflows. |
Class-wise coherence scoring to detect and suppress adversarial data poisoning across distributed federated-learning clients. |
Transformer built from scratch vs. an LSTM/Bahdanau-attention seq2seq baseline on UMC005, with BPE tokenization and BLEU/ROUGE evaluation. |
OCR + ML + NLP Flask app for author identification, sentiment and emotion analysis, grammar checking and plagiarism detection, served over REST APIs. |
🗂️ More projects — click to expand
| Project | Stack | What it does |
|---|---|---|
| Comprehensive Data Pipeline with Apache Airflow | Airflow · Pandas · Matplotlib | DAG-orchestrated ETL and analysis over the Brazilian e-commerce dataset |
| Bagging vs Boosting — Breast Cancer Classification | scikit-learn | Head-to-head of Decision Tree, Bagging, AdaBoost and Random Forest |
| E-commerce Retail Trends & Impact Analysis | Python · Pandas | CLV modelling and what-if pricing analysis on global retail data |
| Iran Airline Analysis | Tableau | 9 sheets and 3 dashboards on air-quality and seasonal trends |
| Console Guitar System | C++ · GUI | Guitar-learning app with user management and a game environment |
| NumPy & Pandas Basics | Jupyter | Teaching notebooks for the data-wrangling fundamentals |
%%{init: {"theme":"base","themeVariables":{"background":"transparent","primaryColor":"#161B22","primaryTextColor":"#E6EDF3","primaryBorderColor":"#1F6FEB","lineColor":"#00D9FF","textColor":"#C9D1D9","fontSize":"14px","cScale0":"#1F6FEB","cScaleLabel0":"#FFFFFF","cScale1":"#00D9FF","cScaleLabel1":"#0D1117","cScale2":"#F0B429","cScaleLabel2":"#0D1117","cScale3":"#8957E5","cScaleLabel3":"#FFFFFF","cScale4":"#3FB950","cScaleLabel4":"#0D1117"}}}%%
timeline
title Journey so far
2022 : BS Data Science @ FAST-NUCES, Islamabad
2023 : Data & AI Engineer — Freelance / Fiverr, international clients (Jun 2023 → Jun 2025)
: Head of Investigations — Baitulnoor
: Donation Officer — Rah-e-Haq (NGO)
2025 : Data Automation Engineering Intern — Tashi Technologies Corp (Jul → Oct 2025)
: Database Manager — Al Muttaqeen Institute (Nov 2025 → May 2026)
: Research — federated learning, XAI, LLM agents
2026 : Data Developer & Analyst — CloudWorks, Texas USA (Jun 2026 →, remote, part-time)
: Senior Machine Learning Automation Engineer — AiPixal, Lahore (Aug 2026 →, on-site, full-time)
: Founder & Artist — Rangrayze (Sufi calligraphy & canvas art, @rangrayze)
: Founder — airRTH (community climate initiative)
: Graduated — BS Data Science
