Looking for a job

Muhammad Usman Qadeer

@muhammadusmanqadeer

Message

I build production LLM, RAG, voice AI, and machine learning systems.

Pakistan

Message

At Traversaal.ai, I fine-tune LLMs and build agentic RAG pipelines that improved domain-specific accuracy by 30%. I also optimize models for mobile and edge devices, reducing inference latency by 40%.

I've built real-time voice AI workflows with Whisper, LLaMA, LiveKit, and WebRTC, and I deploy reliable AI services using FastAPI, Docker, MLflow, and cloud platforms. Previously at Impulxor, I delivered forecasting, demand-planning, ETL, and business intelligence solutions with 85%+ predictive accuracy.

My work also includes satellite-image analysis, land-cover classification, smart water prediction, multilingual LLMs, medical RAG chatbots, and deepfake audio detection. I'm a published researcher in deep learning optimization and spatio-temporal crop classification.

Experience

Work history, roles, and key accomplishments

Current

Senior AI Engineer

Current

Traversaal.ai

Apr 2024 - Present (2 years 5 months)

Fine-tuned LLMs for different industries and used agentic RAG to improve domain-specific accuracy by 30%. Built agentic RAG pipelines using LangChain/LangGraph, enhancing factual reliability for enterprise clients.

LLM Fine tuning RAG Lang Chain LangGraph Quantization Distillation Whisper LLaMA LiveKit WebRTC MLFlow Docker CI CD

AI Engineer

Impulxor

Apr 2021 - Mar 2024 (2 years 11 months)

Built and deployed classical machine learning models using Python and SAS for sales forecasting, demand planning, and marketing analytics, achieving 85%+ predictive accuracy. Designed and automated data ingestion and ETL pipelines using SQL and Python, improving data freshness and reducing manual reporting efforts by 40%.

Python SAS SQL ETL Power BI Google Data Studio Machine Learning Time Series regression Tree Models

AI Software Engineer

Centre for Water Informatics and Technology LUMS

Sep 2018 - Apr 2021 (2 years 7 months)

Analysed 2000+ satellite images using NDVI to monitor crop health and land use across Pakistan. Developed deep learning models for multi-temporal satellite data, achieving 90% accuracy in land cover classification.

NDVI Satellite Imagery Deep Learning Python TensorFlow PyTorch

Education

Degrees, certifications, and relevant coursework

University of Engineering and Technology – Lahore

Masters in Electrical Engineering, Electrical Engineering

Grade: 3.88/4.0

Masters in Electrical Engineering with a thesis on deep learning optimization using Kronecker Product Preconditioning of Stochastic Gradient Descent. Achieved a CGPA of 3.88/4.0.