• 🔬 Research Areas: Transformer architectures, Few-shot learning, Model distillation
• 🛠️ Engineering: End-to-end ML pipelines, Optimization at scale
• 🚀 Innovation: Novel architectures for domain-specific problems
• 📊 Data Science: Transforming raw data into actionable intelligence

• 🤖 Fine-tuning LLMs for specialized industry applications
• 📈 Building scalable ML systems with robust evaluation frameworks
• 🔗 Exploring multi-modal models integrating text, vision and audio
• 🧪 Researching prompt engineering techniques for zero/few-shot learning

Architected a system leveraging domain-adapted LLMs to extract, process, and analyze complex unstructured documents with high precision.

• Implemented parameter-efficient fine-tuning (LoRA) achieving 42% performance boost
• Built custom evaluation framework to measure real-world effectiveness
• Reduced processing time by 68% while improving accuracy by 23%

Tech: PyTorch, Transformers, LangChain, Ray

Developed a multi-modal architecture allowing for classification of previously unseen objects through natural language descriptions.

• Engineered novel embedding alignment mechanism between vision and text spaces
• Created synthetic training data to enhance generalization capability
• Achieved 76% accuracy on zero-shot tasks, surpassing previous approaches

Tech: CLIP, PyTorch, FastAPI, Docker

Built a production-ready abstractive summarization system with domain adaptation capabilities.

• Fine-tuned BART and T5 models on specialized corpora
• Implemented length-controlled generation mechanism
• Achieved 35% ROUGE-L improvement over baseline models

Tech: Transformers, FastAPI, Redis, MLflow

Designed a distributed training infrastructure enabling efficient model training across heterogeneous hardware.

• Implemented adaptive learning rate scheduling based on hardware capabilities
• Built fault-tolerant checkpointing system preserving training progress
• Reduced training time by 78% for large-scale models

Tech: PyTorch Lightning, Kubernetes, Horovod, NVIDIA Triton

• Advanced techniques in model distillation and quantization
• Alignment of large generative models with human values
• MLOps for resource-constrained environments
• Privacy-preserving machine learning methods

• "Deep Learning" by Ian Goodfellow, Yoshua Bengio, Aaron Courville
• "Designing Machine Learning Systems" by Chip Huyen
• "Machine Learning Engineering" by Andriy Burkov
• "Building Machine Learning Pipelines" by Hannes Hapke & Catherine Nelson