Muhammad Ayan Asif
CS @ FAST-NUCES | AI Developer (SkinSense, Grounded) | Former Director at AIESEC
Based in: Islamabad, PakistanMost recently: OC VP Logistics, AIESEC Local Conference, AIESEC
Computer Science student at FAST-NUCES specializing in AI and scalable web systems. Proven track record in leading national-scale platforms for AIESEC and developing high-impact AI projects, from on-device dermatology screening (SkinSense) to brain-inspired neural architectures.
Experience
OC VP Logistics, AIESEC Local Conference
AIESECvolunteer
- Secured venue 18% under budget
- Ran on-site logistics for 50+ delegates
Director, Network Development
AIESECIslamabad, PakistanvolunteerFeb 2025 – Jul 2025
Summary
AIESEC - International NGO advancing UN SDGs. Led development and maintenance of AIESEC in Pakistan's national Member Management web platform, used across multiple local committees. Architected the persistence layer on MongoDB and Firebase, owning schema design and member-record workflows.
Results
- Architected and maintained a centralized member management system that unified data workflows for 11 local committees.
- Scaled the national Member Management platform to serve 1,000+ members across 11 local committees in Pakistan.
MongoDBFirebaseWeb Platform
Project Lead, AIESEC Annual Meeting
AIESECvolunteerFeb 2025 – Apr 2025
Summary
Led a team of 6 for the AIESEC Annual Meeting.
Results
- Negotiated brand partnerships
- Owned end-to-end execution of the annual meeting
Skills
Technical
GitHub
Python
PyTorch
GitMobileNet
C++
DartONNX
ONNX Runtime Web
Data Curation
EfficientNet
FastAPI
MongoDB
FirebaseREST APIs
Progressive Web Apps
FlutterGoogle Gemini API
WSL2 / Ubuntu
Linux CLI
HTML/CSSKnowledge Distillation
Transfer Learning
Android AccessibilityService
Node.js
SQL
C#Git LFS
CLaTeX
Predictive Coding
Model Quantization
JavaScriptx86 Assembly
Fine-tuning
ReactProjects
SkinSense - AI Dermatology Screening for Low-Resource Settings
Final Year Project Leadacademic2025 – Present
Building a 7-class skin-disease classifier targeted at rural clinics in Pakistan, with both an online FastAPI backend and a fully offline on-device inference path. Curated a 12,000-image training set by integrating the DermNet and SkinDiseaseImage corpora; applied class-balanced filtering to drop unreliable label sources. Trained EfficientNet-B3 as the teacher - diagnosed BatchNorm instability and resolved it with full backbone unfreeze from epoch 1 - then knowledge-distilled into a MobileNetV3-Large student, cutting parameters from 11.4M → 3.4M at 71% test accuracy. Redefined deployment from Flutter to a Progressive Web App backed by ONNX Runtime Web after benchmarking RAM and storage constraints on low-end Android devices.
PyTorchFastAPIONNX Runtime WebPWAEfficientNet-B3MobileNetV3-LargeKnowledge Distillation
Urdu Whisper Fine-tuning
personal
Fine-tuned a Whisper-small model specifically for the Urdu language using an LLM-managed pipeline under my supervision.
Whisper-smallLLM
Grounded - Task-Verified Android App Blocker
AI Product Developeracademic2026 – Dec 2026
Shipped an MVP that locks distracting apps until an LLM verifies the user has actually completed a real-world task - differentiates from timer-only competitors. Built the verification engine on the Google Gemini API with multimodal prompting; integrated Android AccessibilityService to block five concurrent apps. Debugged the Flutter/Gradle native build chain end-to-end; produced a Lean Model Canvas validating the value proposition against incumbent focus apps.
FlutterDartAndroid AccessibilityServiceGoogle Gemini APIGradle
Brain-Inspired Neural Architectures
Independent Researcherpersonal2026 – Dec 2026
Designed Model B, a surprise-only propagation network where only prediction-error signals travel between layers; trained 50 epochs to 51.4% test accuracy on the SkinSense 7-class dataset. Engineered a parallel Model A integrating thalamic-style gating, hippocampal memory, and predictive coding modules, with ablations of each. Diagnosed gate-activation failure in Model B and proposed two architectural fixes (reconstruction-loss removal; deeper predictor head) backed by experiment.
PyTorchPredictive Coding
Self-driving car simulator
academic2022 – Dec 2025
Self-driving car simulator in Python trained on real-world driving data.
Python
Pacman and Zuma
academic2022 – Dec 2025
Pacman and Zuma from scratch in x86 Assembly.
x86 Assembly
Daraz clone
academic2022 – Dec 2025
Full-stack Daraz clone in C# with integrated frontend and backend.
C#
Tetris clone
academic2022 – Dec 2025
Tetris clone in C++ without standard data structures.
C++
Media
Education
Bachelor of Science
National University of Computer and Emerging Sciences (FAST-NUCES)Computer ScienceGPA 2.75 / 4Sep 2022 – Jan 2027
Thesis. SkinSense - AI Dermatology Screening for Low-Resource Settings
Coursework
Machine LearningDeep LearningData Structures & AlgorithmsDatabase SystemsDistributed ComputingComputer NetworkingCode OptimizationWeb DevelopmentGame Development
Recognition
Awards
AIESECer of the Quarter
AIESEC in Islamabad
Recognized as a leading team member in Marketing.
Leading Team Member, Marketing
AIESEC in Islamabad
Social impact & activities
Project Lead, AIESEC Annual Meeting
AIESECorganizerFeb 2025 – Apr 2025
Led a team of 6, negotiated brand partnerships, owned end-to-end execution.
Muhammad's twin is AI, it can make mistakes.
More people on Kanvis
Syed shah abdur Rehman
Lahore
Full-Stack Software Engineer & AI Agent Architect
Python · TypeScript · JavaScript · React
kanvis.me/syed-shah-abdur-rehman
Khizer Inayat
Lahore
Business Growth Head, Consultant, Remote and Hybrid Sales and Marketing Executive
Strategic Leadership · Sales Management · Go-To-Market (GTM) Strategy · Business Development
kanvis.me/khizer-inayat
Hafsa Ahad
Lahore
AI Solutions | Technical Sales
Retrieval-Augmented Generation (RAG) · Python · FAISS · LangChain
kanvis.me/hafsa-ahad
Ahmad Zaheer
Lahore
Backend AI Engineer | Full-Stack Developer | Building Scalable AI-Driven Solutions
Urdu · C# · ASP.NET Core · REST APIs
kanvis.me/ahmad-zaheer
