Muhammad Saad Umer

Academic

Career

Verbatim text

Muhammad Saad Umer
Cell: 923005159823 |  Email: saadumer74@gmail.com
LinkedIn: https://www.linkedin.com/in/saad-umer/
Address: BUNGLOW NO E-6, SHAHEEN ROAD PAF ACADEMY ASGHARKHAN, RISALPUR CANTT DISTRICT NOWSHEHRA (KPK)POSTAL CODE
24090 , Risalpur , Pakistan
PROFESSIONAL PROFILE
I am an AI Engineer and Researcher dedicated to bridging the gap between theoretical Deep Learning and production-grade agentic
systems. With a foundation in research from NUST and a deep specialization in Large Language Models (LLMs), my work focuses on
building "thinking" systems that go beyond simple API wrappers.
My expertise lies in architecting high-performance AI systems using a modern stack of FastAPI, React, and Vector Databases . I
specialize in developing advanced Retrieval-Augmented Generation (RAG) pipelines that utilize hybrid search and re-ranking to
solve the "hallucination" problem in document-heavy environments.
Recently, I have focused on:
Model Specialization: Fine-tuning reasoning models like DeepSeek-R1 using Unsloth and QLoRA to align LLMs with
specialized domains, such as Cognitive Behavioral Therapy (CBT).
Agentic Workflows: Designing autonomous agents with LangChain and n8n that can reason through multi-step tasks, utilize
external tools, and self-correct.
Production Deployment: Building asynchronous backends capable of handling low-latency AI inference and streaming
responses for seamless user experiences.
I am driven by the challenge of creating AI that is verifiable, scalable, and capable of autonomous reasoning. I am currently seeking
opportunities to apply my skills in RAG architecture and model fine-tuning to build the next generation of intelligent applications.
EDUCATION
B.E. Software Engineering
School of Electrical Engineering and Computer Science , Islamabad , 3.24 (2026)
INTERNSHIP EXPERIENCE
NUST (College of Aeronautical Engineering)
01-Jun-2024 - 14-Aug-2024
Project: High-Performance Multi-Stream Surveillance System Developed a 2,000+ line production-grade codebase for a surveillance
system capable of processing 20+ simultaneous camera streams. Optimized the system to achieve sub-second mean latency across
all streams, ensuring real-time operational capability. Authored and published research findings in IEEE ICoDT2 2025 based on the
architectural innovations of this system. Engineered a local, noisy audio transcription application using PyQt5 and OpenAI Whisper, optimized for low-resource hardware deployment.
Machine Vision and Intelligent Systems Lab
19-Sep-2024 - 01-Jun-2025
Specialization: LLMs in Cheminformatics Conducted advanced research into the intersection of Large Language Models and
chemical data processing. Published a First Author review paper in a high-impact factor (IF: 12) journal, establishing a
comprehensive framework for LLM applications in Cheminformatics. Collaborated on the development of intelligent vision systems, focusing on the transition from traditional CV to modern transformer-based architectures.
National Aerospace Science & Technology Park (NASTP) Simulator Division
01-Jul-2025 - 01-Aug-2025

AI enrichment