MUHAMMAD ABDULLAH BUTT
ENGINEERING
INTELLIGENCE
“The best model is not the largest one. It is the smallest one that reliably solves the problem deployed on infrastructure that meets real world constraints.”
Senior Full Stack AI Engineer & Researcher. Architecting scalable platforms and parameter-efficient deepfake detection frameworks.
Scroll to ResearchMAPF-Lite
Sole Architect & Lead Researcher
pp 455–474 — First Online: 09 July 2026
Architecture
Frozen CLIP + Whisper backbones, Forgery Signature Gate, SRM/DCT Analysis, Method-Aware Dynamic Prompting, XAI heatmaps. Outperformed AAAI 2025 state-of-the-art (4.4M params) using 99.6% fewer parameters.
In Preparation
Deepfake Detection in the Multimodal Era: A Systematic Review
Lead Author
Surveys multimodal deepfake detection techniques, parameter-efficient architectures, and benchmark datasets, extending MAPF-Lite findings.
estabraq.pk
Estabraq
Fashion E-Commerce PlatformLive production platform delivering sub-2-second load times. GraphQL APIs combined with Redis caching reduced latency by 30%. Automated Docker CI/CD environments cut release cycles by 25%.
mypdfmate.com
MyPDFMate
PDF SaaSLive production SaaS for advanced PDF processing and AI-powered document workflows. AI-driven document analysis and transformation pipelines wrapped in a clean, optimized interface.
khushimotors.com
Khushi Motors
Automotive MarketplaceLive automotive marketplace with comprehensive product catalogue, dynamic filtering, and inquiry management. Production-grade search and secure lead capture.
DeployGenius
AI Chatbot SaaSMulti-tenant AI SaaS with RAG grounding, multi-session agentic workflows, and complete Stripe subscription lifecycle management. Optimized API costs by 25% via token consumption tuning.
LingumedAI
Clinical AI for NursesAI clinical assistant with real-time diagnostic guidance and drug interaction checks. Domain-specific RAG pipeline across medical literature with role-based access control and audit logging.
ForexAI
Mobile Trading & PredictionForex education app with AI market prediction. Fine-tuned XGBoost + Kronos AI ensemble on historical price data with in-app learning curriculum.
Senior Full-Stack AI Engineer
Huzzle.com — London, UK (Remote)
Mar 2026 — Present- AI PLATFORMArchitecting scalable recruitment infrastructure integrating LLM pipelines with reactive frontend applications.
- AGENTIC WORKFLOWSBuilding automated candidate semantic matching, RAG-grounded profile evaluations, and smart parsing.
- HIGH-TRAFFIC B2COptimizing client-facing user experiences to maintain sub-100ms response targets under peak traffic.
Founder & Lead Architect
Neural Stack — Remote
2026 — Present- AI STRATEGYDirecting technical roadmaps, enterprise AI blueprints, and high-performance system designs.
- ENTERPRISE RAGDelivering private LLM deployments, pgvector knowledge grounding, and fine-tuned agent architectures.
- SYSTEM SCALINGAdvising engineering teams on cost-effective model routing and distributed cloud infrastructure.
Full Stack Engineer
MAQ Enterprises Ltd. — Rawalpindi, Pakistan
Jan 2023 — Feb 2026- 40% REDUCTIONPage load time cut via Next.js SSR optimization, dynamic code splitting, and bundle trimming.
- 35% SPEEDUPDatabase retrieval acceleration achieved through PostgreSQL relational schema and MongoDB index tuning.
- 30% FASTERRelease cycle time cut by deploying automated Jenkins and Docker CI/CD pipelines.
- 25% FEWER BUGSProduction bug incidence reduced using rigorous Cypress end-to-end and Jest unit test suites.
- TERRAFORM IACProvisioned automated cloud infrastructure and reproducible environments with Infrastructure as Code.
AI Researcher
GIFT University — Gujranwala, Pakistan
2024 — 2026- SPRINGER NATUREConceived, authored, and published MAPF-Lite at FLINS-ISKE 2026 (Springer Nature proceedings).
- 99.6% PARAMETER CUTReduced trainable parameters from 4.4M+ down to 0.566M using PEFT while matching SOTA accuracy.
- 30% LATENCY CUTEngineered lightweight attention mechanisms and frequency-domain filters for real-time edge detection.
Full-Stack Developer
Estabraq — Pakistan
2025 — May 2026- SUB-2S LOADSArchitected high-throughput fashion platform with Next.js App Router, Prisma ORM, and Supabase.
- 30% LATENCY CUTIntegrated GraphQL APIs, Redis caching layers, and edge CDN asset optimization.
- 25% FASTERAccelerated release cycles and automated zero-downtime rollouts via Docker CI/CD pipelines.
- Python
- TypeScript
- JavaScript
- Java
- C++
- PHP
- LangChain
- LlamaIndex
- OpenAI
- Anthropic Claude
- Hugging Face
- RAG Pipelines
- pgvector
- Pinecone
- PyTorch
- TensorFlow
- Keras
- Scikit-Learn
- MLflow
- Ray
- CUDA
- Vision Transformers
- CLIP
- Whisper
- XGBoost
- Kronos AI
- PEFT
- Next.js
- React
- Angular
- Vue.js
- Svelte
- React Native
- Expo
- Redux
- Webpack
- Vite
- Tailwind CSS
- Material UI
- HTML
- CSS
- Node.js
- Express
- NestJS
- FastAPI
- Django
- Spring Boot
- ASP.NET
- Go Fiber
- Prisma ORM
- GraphQL
- REST APIs
- gRPC
- Apollo
- PostgreSQL
- MySQL
- MongoDB
- Redis
- Cassandra
- Elasticsearch
- DynamoDB
- Supabase
- Oracle DB
- AWS
- GCP
- Azure
- Oracle Cloud
- Docker
- Kubernetes
- Terraform
- Jenkins
- GitHub Actions
- GitLab CI
- Ansible
- Prometheus
- Grafana
- Nginx
- Apache Kafka
- RabbitMQ
- CI/CD
- Vercel
- Railway
Bachelor of Science in Computer Science
GIFT University, Gujranwala, Pakistan
Sep 2022 — Mar 2026MAPF-Lite — published at FLINS-ISKE 2026 and included in Springer Nature proceedings volume.
- 01AMD Developer Hackathon Act II Participant (Team Neural Stack)
- 02Harvard HSIL Health Systems Hackathon Participant (GIFT University)
- 03MAPF-Lite included in Springer Nature proceedings volume (FLINS-ISKE 2026)
START A
CONVERSATION.
Available for senior engineering leadership, AI systems research, or technical advisory. Direct inquiries welcome.
“The best model is not the largest one. It is the smallest one that reliably solves the problem deployed on infrastructure that meets real world constraints.”
Muhammad Abdullah Butt — MAPF-Lite (FLINS-ISKE 2026)