Muhammad Moiz

Atlanta, USA · Remote

• Own backend architecture end-to-end for a live, production mobile app, from system design through deployment.
• Lead sprint execution and coordinate delivery for the engineering team alongside hands-on individual contributor work.

Estonia · Remote

• Built a production agentic system automating an entire email-based ship buying/selling workflow for a maritime client, replacing a fully manual process.
• Architected LangGraph multi-agent pipelines with Langfuse observability and integrated vector search for retrieval-augmented context.

UK · Remote

• Owned architecture decisions and led client meetings for two Flutter/Firebase consumer apps: a trail-booking app (Encounter Adventures) and a merchant-discounts app (Infinity Club).
• Shipped both apps to the Play Store and App Store.

Lahore · On-site

• Designed and built a food-rescue platform (surplus food pickup/redistribution) as the internship's capstone project.
• Conducted market research and presented a business case to company leadership evaluating the platform's viability beyond the internship.

React/Next.js, FastAPI, PostgreSQL, OCR, LLM

• Architecting a tax-reconciliation platform for Pakistani taxpayers with a hybrid AI/deterministic pipeline: AI handles fuzzy tasks (OCR, document classification) while all tax-figure arithmetic stays fully deterministic to eliminate hallucination risk.
• Designing a wealth-reconciliation diagnosis engine using bounded subset-sum search and evidence-ranked pattern detectors to explain discrepancies rather than just flag them.
• Targeting ≥85% document field-extraction accuracy, ≥80% transaction-classification accuracy, and a ≥50% reduction in manual reconciliation effort versus a manual baseline, measured on synthetic ground-truth datasets.

Python, TensorFlow/Keras, CNN

• Compared 7 deep learning architectures for facial emotion classification on FER-2013; best model (CNN with data augmentation) reached 58.3% test accuracy, approaching the ~65% human-level baseline, backed by a full ablation study.

Python, FastAPI, React, GenAI

• Automated resume classification system reaching 96.3% accuracy across 18 tech roles, using dual-vectorization (TF-IDF) and synthetic data augmentation (Llama-3.1) to fix severe class imbalance in the training data.

Full Stack, LLM

• CV-matching tool for Mitacs Globalink applicants, cutting the roughly 40-hour manual review of 3,300+ available research projects down to a scored, ranked shortlist. Built with a peer; contributed domain logic and testing.

Next.js, Supabase, Stripe

• Coursera-style platform built in 8 hours with a 3-person team: course listings with Stripe payments end to end, live lecture transcription, and in-course chat.

Flutter, Agora, Firebase

• Telemedicine platform for live video consultations, doctor-patient appointment booking, and digital prescriptions; led development end-to-end as a university Software Engineering course project.

Next.js, Full Stack

• Real-time carpooling platform matching university commuters to cut transportation costs and reduce environmental impact.

C++, Qt, Python, Flask

• Sentiment analysis and Trie-based autocomplete tool combining a C++/Qt desktop interface with a Python/Flask backend for real-time text prediction and mood detection; led development as an AI course project.

C, Operating Systems

• Added a custom priority-based process scheduler to xv6 (MIT's teaching OS), including new syscalls and kernel-level scheduler modifications, backed by an original test suite.

Chadar — NUST student-led community service club

• Organized clothes donation drives for underprivileged communities, covering both collection and distribution.

Chadar

• Helped organize and distribute Iftar meals during Ramadan.

Summer School Bootcamp — NUST partner program

• Tutored 12th-grade students in an intensive pre-university bootcamp covering 21st-century skills and practical tools ahead of their transition to NUST.

NUST Community Services Club (NCSC)

• Regular donor at semesterly campus blood donation drives.