Dynamic AI engineer turning complex challenges into intelligent solutions. Specializing in RAG pipelines, LLM architectures, agentic AI systems, and production-grade ML applications.
I'm a results-driven AI engineer with expertise in building end-to-end intelligent systems. From designing multi-agent RAG architectures to deploying real-time computer vision pipelines, I thrive on transforming complex problems into elegant, production-ready solutions.
Currently working at doAZ (Seoul, South Korea) as an AI Engineer, building enterprise-scale AI platforms for industrial safety management and real estate intelligence.
My research is focused on linguistic AI for low-resource languages — with a published paper on Hate Speech Detection for Roman Urdu.
Leading end-to-end AI platform development for enterprise clients. Building multi-module intelligent systems integrating RAG pipelines, tool-calling agents, and FastAPI microservices. Designing bilingual (Korean/English) chatbots, real-time streaming architectures, and large-scale document processing systems for industrial and real estate sectors.
Collaborated with the AI team to develop and enhance existing AI models. Researched new AI technologies including RAG architectures, vector databases, and LLM integrations. Contributed to the development of production-grade AI applications for Korean enterprise clients.
Multi-module AI platform for industrial safety management. Integrates RAG, tool-calling agents, and FastAPI microservices powering Risk Assessment, Multi-Prompt AI, Ask Doosan AI, and Legal Compliance modules. Bilingual Korean/English chatbot with real-time streaming.
Comprehensive modular AI system with 4 RAG pipelines and 4 agents for legal queries, document QA, multi-document summarization, construction drawing analysis, and Excel automation. Supports PDF, DOCX, XLSX, and image extraction with multi-language support.
Intelligent mock interview system based on the user's CV. Uses Gemini Flash 2.5 to generate interview questions, Deepgram TTS for speech output, and Deepgram STT to transcribe responses. Provides performance ratings out of 10 after the interview session.
Agentic AI system designed to automate and augment HR workflows. Handles candidate screening, question generation, and evaluation processes using advanced LLM reasoning and tool-use capabilities.
Real-time construction safety gear detection using YOLOv8. Monitors workers for helmets, goggles, gloves, vests, and boots via webcam. Trained on a custom-annotated dataset and deployed as a Streamlit web application with image/video upload support.
Robust and linguistically-aware hate speech detection system for Roman Urdu. FYP and research project involving custom NLP pipeline, dataset curation, and model training. Tied to published research paper on Google Scholar.
Deep learning image classification model distinguishing between bikes and cars. Implements CNN-based architecture with data augmentation and transfer learning techniques for high-accuracy vehicle classification.
Machine learning model predicting house prices for rent and sale in Lahore. Covers full data science pipeline including EDA, feature engineering, and regression modeling on real estate data from Pakistan's second-largest city.
Zero-shot TTS system that generates speech in the same voice as input audio. Uses MaskGCT for TTS conversion and Whisper for accurate transcription, enabling seamless voice cloning while maintaining naturalness and clarity.
A comprehensive NLP research contribution developing a robust hate speech detection system tailored to Roman Urdu — a low-resource, code-mixed language. The system leverages linguistically-aware modeling techniques to identify and classify hateful content in informal digital text, addressing a critical gap in multilingual AI safety research.
I'm open to exciting AI engineering opportunities, research collaborations, and freelance projects. Whether you need a RAG system, a computer vision solution, or a full AI pipeline — let's talk.
Send a Message