MuhammadAfaq Jamshaid
Master's student in Artificial Intelligence at Brandenburgische Technische Universität, passionate about advancing AI/ML research and open-source development. Currently working as a HiWi (Research Assistant) contributing to the open-source tool pandapipes. Hands-on experience in prompt engineering and LLM optimization with independent ML projects experience in startup environments.
Technical Skills
Programming Languages
Frameworks
Libraries
Tools & Platforms
Areas of Expertise
Professional Experience
My journey in AI/ML research and development
Research Assistant (Hiwi)
Brandenburgische Technische Universität Cottbus
- Utilized Python and the pandapipes library to create and simulate pipeline network topologies for CFD research, analyzing simulation data and visualizing network configurations.
- Contributed to the open-source development of the pandapipes library, enhancing its functionality and improving simulation capabilities.
Artificial Intelligence Intern
LOGEsoft
- Developed a customized AI pipeline combining neural networks and reinforcement learning to predict fuel compositions with desired properties, enhancing predictive capabilities through end-to-end implementation from data acquisition to model integration.
- Automated predictions with Python and Perl scripts, leveraging a hybrid AI approach for efficient and accurate fuel composition predictions.
LLM Prompt Engineer
Independent Collaboration with Industry Professional
- Created and refined over 1000 training data scripts for LLMs, optimizing model performance through supervised fine-tuning and data post-processing.
- Improved AI response accuracy through RLHF by evaluating and correcting AI responses based on rubrics assessing accuracy, instruction following, and efficiency.
- Crafted scenario-based prompts for LLM evaluation and developed test cases to systematically assess model responses, verifying whether they met the defined requirements.
Featured Projects
SMENTRY - Access Control System
Contributed to the development of an access control system utilizing computer vision to detect and authorize resident vehicles' number plates, which resulted in a 25% increase in entry processing efficiency.
Key Achievements:
- Trained a custom YOLOv8 model and performed OCR, improving detection accuracy and reducing manual entry errors by 30%
- Integrated the trained model into the web portal front end, streamlining entry management and reducing processing time by 20%
German Text Classification with DistilBERT
Developed a robust text classification solution using DistilBERT, categorizing German phrases within a given dataset. Deployed the model using FastAPI, with the entire project containerized using Docker.
Key Achievements:
- Successfully deployed machine learning model with FastAPI for production use
- Containerized entire application using Docker for scalable deployment
- Implemented efficient German text processing pipeline
Pandapipes Open Source Contributions
Active contributor to the pandapipes library, an open-source Python tool for modeling and analyzing pipe networks.
Key Achievements:
- Enhanced library functionality for CFD research applications
- Improved simulation capabilities and network topology creation
- Contributed to community-driven development and documentation
Hybrid AI Pipeline for Fuel Composition Prediction
AI-driven pipeline combining supervised and reinforcement learning to optimize complex fuel blends across multiple targets, automating the full data workflow and reducing simulation costs.
Key Achievements:
- Explored millions of blend combinations efficiently using a modular pipeline with feature-weighted scoring and zoom-in data generation, avoiding local minima and improving optimization accuracy.
- Accelerated optimization cycles by integrating supervised and reinforcement learning in a continuous improvement loop.
ClarityQA Chat App
Streamlit-based chat app with document upload, vector search, and RAG-powered responses using Mistral LLM, enabling fast, context-aware conversations with the document.
Key Achievements:
- Delivered accurate, context-aware Q&A over uploaded documents using RAG with Mistral LLM and Chroma DB.
- Improved performance and reliability with persistent vector storage, efficient session management, and resource cleanup.
Education
Master of Science in Artificial Intelligence
Brandenburgische Technische Universität
Cottbus, Germany
Advanced studies in machine learning, computer vision, neural networks, and AI research methodologies. Focus on practical applications and cutting-edge AI technologies.
Bachelor of Science in Computer Science
FAST NUCES
Pakistan
Comprehensive foundation in computer science fundamentals, programming, algorithms, data structures, and software engineering principles.
Certifications
The Power of Statistics
Google (Coursera)
Go Beyond the Numbers: Translate Data into Insights
Google (Coursera)
Get Started with Python
Google (Coursera)
Foundations of Data Science
Google (Coursera)
Introduction to TensorFlow for AI, ML, and DL
DeepLearning.AI (Coursera)
Let's Connect
Get in Touch
Feel free to reach out through any of these channels
Professional networking
afaq.jamshaid123@gmail.com
© 2024 Muhammad Afaq Jamshaid. Built with React, TypeScript, and Tailwind CSS.