Open to work — Pakistan
Machine Learning · MLOps · Pakistan
I build machine-learning systems the slow way - from notebook to production, properly.
I’m a data-science student in Pakistan who builds machine-learning systems end to end - training, experiment tracking, containerization, and the REST API at the far side. The glamorous part is the model. I take care of everything around it.
Right now I’m finishing my bachelor’s at Riphah International University and doing ML work for an economics PhD thesis. In between, I look for hard problems with unglamorous middles - scheduling around load-shedding, segmentation under noisy scans, ensembles on a competition clock.
Tools change. The discipline doesn’t.
Autonomous build pipeline
Seven phases — research, design, architecture, backend, review, security, deploy — each running in a fresh session and handing off through files on disk. Every phase loads a domain knowledge vault rather than trusting the model's memory; the frontend context alone distills 152 open-source design systems.
Python · LLM Agents · Context Engineering · Obsidian
Production planning around load-shedding
A proof of concept for scheduling a textile mill's production orders around LESCO's off-peak tariffs. FastAPI backend, a React calendar view, APScheduler scraping load-shedding notices, and a greedy cost-sorting heuristic — containerized with Docker Compose and deployed on Railway.
FastAPI · React · PostgreSQL · Docker Compose · Railway
ResNet-34 pipeline
A segmentation model with a ResNet-34 encoder, trained in PyTorch on Kaggle GPUs with Albumentations handling the preprocessing. 0.92 Dice on the validation set.
PyTorch · ResNet-34 · Albumentations
Labour-market economics, PhD thesis
Machine-learning support for a doctorate on AI adoption in Pakistan's IT sector: data cleaning, Random Forest and Logistic Regression models, and the evidence a thesis committee asks for — SHAP values, ROC curves, regression diagnostics.
Scikit-learn · Statsmodels · SHAP · Pandas
Stacking ensemble, ML competition
A full stacking ensemble built under competition time pressure — LightGBM, XGBoost, CatBoost, HistGradientBoosting, TabNet, and a PyTorch MLP, tuned with Optuna. My first taste of end-to-end competition ML, and not the last.
LightGBM · XGBoost · CatBoost · TabNet · Optuna
iii. the vortex
BS Data Science — Riphah International University, Lahore (expected 2028)
Linear algebra · Probability & statistics · Data structures & algorithms · Databases · Analysis of algorithms
PyTorch · TensorFlow · Machine Learning · AI Agents
Zero To Mastery Academy, 2024–25 — and IoT Fundamentals, Cisco Networking Academy, 2024
Open to junior MLOps and ML engineering roles in Pakistan - and to interesting problems anywhere.
dawood.ml@outlook.com