Machine Learning Engineer | NLP & AI Researcher
I specialize in fine-tuning large language models (LLMs), multi-GPU training, and developing AI-driven applications. My expertise spans NLP, speech AI, vision AI, and agent-based systems, leveraging advanced techniques like RLHF, DPO, KTO, GRPO, and model distillation.
I deploy and optimize LLMs using Hugging Face TGI, vLLM, and sglang, ensuring efficient, scalable AI solutions for real-world applications.
🔬 Expertise: LLM fine-tuning, Speech AI, Vision AI, Model Distillation
⚡ Currently working on: Fine-tuning Qwen2.5-72B for multilingual AI applications
🏆 Passionate about: NLP, RAG (Retrieval-Augmented Generation), AI agents
🤝 Open to collaborations in AI research, multi-modal models, and cloud-based AI deployment
- Fine-tuned large models using SFT, RLHF, DPO, Online DPO, KTO, GRPO
- Alzheimer's disease prediction using CNN
- Pashto poetry generation using pre-trained transformers
- Worked with GANs, VAEs, and Diffusion models
- Synthetic dataset generation
- Developed AI-driven assistants with CrewAI, LangChain, LangGraph
- Built real-time speech AI solutions (Whisper, TTS)
- Deployed LLMs on AWS, SageMaker, and on-prem GPUs
- Integrated Roboflow for vision AI solutions
- MEDICAL Reasnoning llm
LLM Training & Fine-Tuning: PyTorch, TensorFlow, Transformers, Datasets, JAX, DeepSpeed, FSDP, LoRA
AI Agents & RAG: CrewAI, LangChain, LangGraph, LlamaIndex
Cloud & Deployment: AWS, SageMaker, VLLM,SGLang, Triton, Kubernetes
Speech & Vision AI: Whisper, TTS, Roboflow, YOLO
Programming: Python, C++, JavaScript, Bash
MLOps & DevOps: MLflow, Docker, CI/CD
- Generative AI with Large Language Models
- Python Essentials for MLOps
- Build, Train and Deploy ML Models with Keras on Google Cloud
- Getting Started with AWS Machine Learning
- Introduction to TensorFlow for Artificial Intelligence, Machine Learning, and Deep Learning
- Accelerating Long-Sequence Transformers with Ring vs. Striped Attention on Multiple GPUs
- Detailed breakdown of the math formulas, dataset examples, and training methods from the DeepSeek R1 paper
- Unlocking Efficiency: A Deep Dive into Medical Model Fine-Tuning with Unsloth, TRL and PEFT
- OpenChat 3.5: A Deep Dive into SFT Fine-Tuning with Unsloth
