I'm an AI/Full Stack engineer building Goniaa, a fashion e-commerce ecosystem I designed, built, and shipped solo — a public storefront, an AI-powered brand management dashboard, and a native mobile app, all in production with real users and real data.

• BS Artificial Intelligence @ Hazara University (Expected June 2027)
• Currently building Goniaa end-to-end: storefront → AI ops dashboard → mobile app
• Comfortable across the stack: Next.js/TypeScript frontends, Supabase/Postgres backends, LLM-powered internal tools, and React Native mobile
• Interested in applied ML, agentic tooling, and building products that actually ship — not just notebooks
• Open to remote roles with US/UK startups

• Shipping v2 of the Goniaa AI Brand Manager — an 11-tab operations dashboard that uses an LLM (Groq / Llama 3.3 70B) to help run a real e-commerce brand
• Stabilizing the Goniaa mobile app (React Native + Expo SDK 54) ahead of a public release
• Studying applied system design — moving from "it works" to architecture that scales
• Preparing for AI/Full Stack internships and junior roles at product-focused startups

Founder & Solo Developer · Live in production

A full ecosystem, not a demo: a public storefront customers actually buy from, an internal AI-driven dashboard that runs the brand's operations, and a companion mobile app — architected, built, and shipped end-to-end by one person.

Engineering highlights

• Designed and implemented an 11-tab dashboard from a blank repo to production, including auth, data modeling, and an LLM integration layer
• Diagnosed and resolved a production-blocking React error (recharts #130) by replacing the charting layer with custom CSS-based charts
• Root-caused a recurring CSSStyleDeclaration crash traced to JSX spread operators in style props, and eliminated it by refactoring to reusable styled components
• Fixed Supabase auth misconfiguration and replaced deprecated dependencies to unblock the build

Live Site · Dashboard Demo Video · Mobile APK · Case Study Repo

Deep learning pipeline classifying red blood cell abnormalities on the AneRBC dataset. Benchmarked custom CNNs against transfer learning (MobileNetV2, ResNet18, DenseNet121), reaching 82.7% accuracy with DenseNet121, and added Grad-CAM explainability to visualize model decisions.

Stack: PyTorch · Transfer Learning · Grad-CAM Repo →

DQN and tabular Q-learning agents trained on classic control environments (CartPole, Lunar Lander, FrozenLake), covering exploration strategies, reward shaping, and convergence analysis.

Stack: Python · PyTorch/Gymnasium Repo →

A set of NLP models covering classification (fake news, spam) and text summarization, built to compare classical ML pipelines against transformer-based approaches.

Stack: Python · scikit-learn · NLP Repo →