Muhammad Asim

AI Engineer specializing in LLM solutions, RAG pipelines, and Computer Vision.

Based in: Islamabad, PakistanMost recently: AI Engineer, Arfa Tech

Hire me

Computer Science student at NUST with professional experience developing production-ready AI applications and YOLO-based computer vision models. He has a proven track record in building explainable AI systems and multi-agent research assistants using modern frameworks like LangGraph and PyTorch.

Experience

AI Engineer

Arfa TechRawalpindi, PakistanworkJul 2026 – Sep 2026

Summary

Developed AI/LLM solutions and production-ready AI applications, focusing on reliable workflows, model integration, and scalable system design. Optimized AI workflows and applications by improving retrieval, prompting, processing pipelines, and overall system efficiency. Developed React-based AI applications and implemented backend middleware, API gateways, authentication and security controls to support reliable and secure system integration.

Results

  • Improved pipeline performance and retrieval quality by reducing unnecessary retrieval and processing steps.
  • Made system responses faster and more consistent through workflow optimization.
AILLMReactAPI gatewaysAuthentication

AI and Computer Vision Intern

Forward SportsSialkot, PakistaninternshipJun 2025 – Aug 2025

Summary

Built and prepared annotated datasets for YOLO-based object detection, supporting real-time computer vision applications for factory automation. Trained and optimized YOLO deep learning models for automated visual inspection and production-grade object detection. Developed computer vision pipelines for real-time quality inspection, reducing reliance on manual inspection through automated AI-based detection.

What I did

  • Deployed the YOLO-based inspection system into production.

Results

  • Achieved 95.7 percent accuracy for the automated quality inspection system.
  • Processed three patches per second compared to 2 to 5 seconds for manual labor.
YOLOComputer VisionDeep Learning

Skills

Technical

PyTorchPyTorch
RAG Pipelines
ReactReact
PythonPython
LangGraph
OpenCVOpenCV
HTML/CSSHTML/CSS
Qdrant
JavaScriptJavaScript
FastAPIFastAPI
GitGit
LangChain
scikit-learn
PostgreSQLPostgreSQL
Java
CC
SQLSQL
Next.jsNext.js
Node.jsNode.js
Express.jsExpress.js
DockerDocker
TensorFlowTensorFlow
FAISS
FirebaseFirebase
Make.com
FlutterFlutter
YOLO
TypeScriptTypeScript
Machine Learning
NLP
Fine-tuning

Projects

WAVESAGE (FYP) - Explainable EEG AI

Lead Developeracademic
Built a responsive React and TypeScript web application for EEG micro-event localization, featuring interactive dashboards and real-time data visualizers powered by CNNs and wavelet decomposition. Integrated SHAP-based feature attributions into modular frontend UI components to render intuitive model explanations and playback controls, achieving 64.2% F1-score and 79% recall on clinical datasets.
PyTorchReactCNNSHAPTypeScriptWavelet Decomposition

NUST LLM Chatbot

Developerpersonal
Developed a citation-grounded RAG pipeline for NUST policy documents using Groq, Jina embeddings, and Qdrant hybrid retrieval with reranking. Evaluated retrieval quality using nDCG and Recall@K, and deployed the containerized application via a Dockerized Streamlit interface.
GroqJinaQdrantRAGDockerStreamlit

Multi-Agent Research Assistant

Developerpersonal
Developed a multi-agent research assistant using LangGraph with automated planning, web research, reflection, and synthesis workflows. Integrated Groq and Tavily to coordinate LLM reasoning with external information retrieval and structured research outputs. Implemented rate-limit backoff, circuit-breaker logic, and semantic gibberish detection to improve reliability and reduce hallucination risks.
LangGraphGroqTavilyLLM

Resume Analyzer

Developerpersonal
Built a web-based resume analyzer application with a JavaScript frontend and an interactive interface for file uploads, job descriptions, and match metric visualization. Engineered semantic matching logic using PyTorch and Sentence Transformers embeddings with cosine similarity to evaluate candidate relevance, achieving 78% matching accuracy.
PythonJavaScriptNLPPyTorchSentence Transformers

Sentiment Analysis Project

personal
Developed a project where text was classified based on sentiment, involving the complete ML workflow from preprocessing to model evaluation.

Education

Bachelor of Science

National University of Sciences and Technology (NUST)Computer ScienceSep 2022 – May 2026

Thesis. WAVESAGE - Explainable EEG AI

Hire MuhammadReach out about a role, a contract or a conversation.For recruiters

Muhammad's twin is AI, it can make mistakes.

More people on Kanvis

Browse more peopleCreate your own Kanvis