Skip to content

based in Islamabad · working with Sydney · open to interesting problems


AI Engineer building LLM agents

I build LLM agents, voice assistants, and computer-vision systems, from research prototypes to production-deployed applications. Passionate about merging AI and hardware for next-gen computing.

GitHubLinkedInInstagram

About

Muhammad Mustafa

I'm an AI Engineer working full-time with Beresfords Wealth Management (Australia), where I build and deploy AI solutions across business processes. I hold a Bachelor's in Electrical & Electronics Engineering from NUST Islamabad, specializing in AI and hardware design. My work spans the full AI stack: training and fine-tuning ML models, building LLM-powered agents, voice agents and chatbots, computer-vision pipelines, and shipping them inside full-stack web applications. I've also worked at the hardware end of the field, from FPGA-based deep learning to hardware-Trojan detection with Graph Neural Networks.

When I'm not shipping AI systems, you'll find me deep in One Piece lore or dodging bosses in Elden Ring. 🎮

0+
years building AI systems
0+
projects shipped
0
published paper

Experience

  1. Beresfords Wealth Management

    Aug 2025 – Present · Sydney, NSW · Hybrid
    AI Innovation SpecialistJul 2026 – Present · Full-time
    AI ConsultantAug 2025 – Jun 2026 · Part-time
    • Guiding and implementing AI solutions across business processes to improve efficiency, accuracy, and client experience.
    • Collaborating with management to identify where AI can deliver measurable value and streamline workflows.
    • Developing and integrating AI tools for data-driven decision-making, predictive analytics, and process automation.
    • Providing strategic consultation on AI adoption, feasibility assessment, and responsible deployment.

  2. Salik Labs

    Jun 2025 – Jun 2026 · Islamabad · On-site

    AI Engineer

    • Worked on building a startup in the hiring and recruitment space.
    • Worked with multiple founders to take products from zero to one.
    • Built AI agents, voice agents, AI automations, and chatbots for various startups.

  3. AI Data House

    Feb 2025 – Apr 2025 · Islamabad · On-site

    Junior AI Engineer

    • Developed a handball game statistics program using computer vision for real-time sports analytics.
    • Created, tested, and fine-tuned LLMs for diverse chatbot applications.
    • Integrated machine-learning backends into websites to enhance functionality and UX.

  4. ROMI Lab, NUST SEECS

    Apr 2024 – May 2025 · Islamabad · Hybrid

    Researcher

    • Hardware-security research in collaboration with professors at Tennessee Tech, USA.
    • Detected Trojans in IP cores at the SoC development stage; basis of the NUST Final Year Project.
    • Converted Verilog into graph structures capturing IP module hierarchy, then classified them with Graph Neural Networks.

  5. NCAI TUKL Deep Learning Lab, NUST

    Jun 2023 – Feb 2024 · Islamabad · On-site

    Research Intern

    • Worked on several ML architectures, focused on Vision Transformers and ViT-based object detection.
    • Reduced the computation required to run a ViT so it could run on edge devices.
    • Goal: a Vision Transformer for an ADAS system used in autonomous vehicles.
Also:Google Developer Student Clubs · HR Executive (Sep 2023 – Jun 2024)Hack Club NUST · Event Coordinator (Dec 2021 – Jun 2023)

Projects

  • Flask REST API wired to LangChain, OpenAI GPT-4o, and ChromaDB for document processing.
  • Retrieval-Augmented Generation over uploaded PDF contracts: detailed summaries plus free-form Q&A.
  • Responsive React frontend for uploading documents, viewing summaries, and chatting with the assistant.
FlaskReactLangChainGPT-4oChromaDBRAG
Read the case study Live Repo
  • Dynamically manages conversation flows based on user input and graph state.
  • Real-time client-server communication over WebSockets with instant state sync.
  • Frontend visualizes conversation history and the structured JSON state live.
LangGraphLangChainLocal LLMWebSockets
Read the case study Live Repo
  • Voice interaction and camera understanding with zero cloud dependency: everything runs on-device.
  • Built for the trail: works where there is no connectivity at all.
  • AI on the edge, where the hardware and ML sides of the stack meet.
PythonRaspberry Pi 5Edge AISpeechVision
Read the case study Repo
  • Generates role-specific interview questions from the job description and the candidate's CV.
  • Evaluates answers automatically and scores candidates against the role.
  • End-to-end hiring assistant: from posting to structured candidate evaluation.
PythonLLMsPrompt EngineeringEvaluation
Repo
  • User-based and item-based collaborative filtering on the MovieLens 20M dataset.
  • Federated learning shares model updates only, never raw user data.
  • Tackled scalability and cold-start challenges of large-scale recommenders.
Federated LearningCollaborative FilteringPython
Repo
  • Compared Deep Q-Network, CNN, and MobileNet-V2 approaches on the same environment.
  • Best results with a Deep Q-Neural Network after reward shaping and frame stacking.
Reinforcement LearningDQNPyTorchOpenAI Gym
Repo
  • Converted 1-D time-series signals into 2-D image-like data and trained a CNN to separate noisy from clean signals.
  • Deployed on a XILINX FPGA using High-Level Synthesis for optimized real-time inference.
CNNXILINX FPGAHLSPyTorch
  • Responsive UI with an integrated patient-ticket workflow and expense tracking.
  • MERN stack, deployed on Firebase.
MongoDBExpressReactNode.jsFirebase
Live Repo
  • Supports arithmetic, branching, and looping; simulated end-to-end in Proteus.
  • Digital design from the instruction set up: registers, ALU, and control logic.
Digital DesignProteus
Repo

More on GitHub

Full profile on GitHub

Skills

Languages

Python
C/C++
JavaScript
TypeScript
Verilog
AVR Assembly
SQL
LaTeX

AI / ML

LLM Agents
Voice Agents
RAG
Model Fine-tuning
Computer Vision
Reinforcement Learning
Transformers / ViT
Graph Neural Networks
Federated Learning
AI Automation

Frameworks

PyTorch
TensorFlow
Keras
LangChain
LangGraph
Hugging Face
OpenCV
Scikit-Learn
NumPy
Pandas

Web / Full-Stack

React
Next.js
Node.js
Flask
FastAPI
MERN
WebSockets

Cloud / DevOps

Docker
AWS
Google Cloud
Vercel
Render
Firebase
Git

Hardware / Embedded

FPGA (XILINX, HLS)
Digital System Design
Raspberry Pi
Arduino
Embedded Systems

Credentials

National University of Sciences and Technology (NUST), Islamabad

B.E. Electrical & Electronics Engineering

Nov 2021 – Jun 2025

  • CGPA 3.5
  • Final Year Project: Hardware Trojan Detection using Graph Neural Networks
  • Coursework: AI, Computer Vision, NLP, Embedded Systems, Digital System Design, DSA, DSP

Cadet College Hasanabdal

FSc Pre-Engineering

Apr 2019 – Jul 2021

  • Abdalian '63
PUBLICATION
“Speaker Recognition: A Comparative Analysis Between Deep Learning and Non-Deep Learning Methodologies”

Asian Bulletin of Big Data Management

Muhammad Mustafa, Ahmad Faisal Mirza, Zoha Ahmed, Sakhi Usman Akbar

Certifications

  • Neural Networks and Deep LearningDeepLearning.AI
  • Structuring Machine Learning ProjectsDeepLearning.AI
  • Sequence ModelsDeepLearning.AI
  • Supervised ML: Regression and ClassificationDeepLearning.AI
  • Unsupervised Learning, Recommenders, RLDeepLearning.AI

IELTS Band 8 · English (professional) · Urdu (native)

Contact

Let’s build
something.

Have an AI product to ship, a model to tame, or just want to talk shop? My inbox is open.

GitHubLinkedInInstagram