Yash Ghogre
AI Engineer
AI Engineer transforming complex problems into intelligent, scalable solutions—from custom ML models to integrated LLM applications.
About Me
I'm an AI Engineer driven by a deep curiosity to understand how complex systems work, not just that they do. This curiosity has led me to build foundational LLM architectures like LLaMA 2 and GPT-2 from scratch just to see their inner workings. I apply this same "from-the-ground-up" mindset to solve practical problems, whether I'm architecting a scalable memory framework or optimizing code to win a GPU-accelerated computing codeathon. I thrive on bridging the gap between deep theory and real-world application, building intelligent solutions that are robust and highly efficient.
Featured Projects
Mem1: Memory Framework for LLMs
Independently developed a scalable memory framework for LLMs and autonomous agents based on the Mem0 research paper, engineering a multi-component retrieval pipeline and a CLI assistant.
Core LLM Architecture (LLaMA 2 & GPT-2)
Engineered complete, from-scratch PyTorch implementations of LLaMA 2 (7B) and GPT-2 (124M), demonstrating deep proficiency in modern transformer design and components like RoPE, GQA, and KV Caching.
Autograd Engine from Scratch
Designed and implemented a Python-based automatic differentiation engine, supporting dynamic computation graphs and diverse tensor operations, improving computational efficiency by 30%.
Tech Stack
Programming Languages
Frameworks/Libraries
Databases
Cloud & Tools
Work Experience
Turbo ML (Puch AI)
AI Engineering Intern
Enhanced user experience by integrating robust web-search functionality into a WhatsApp Chatbot, increasing information retrieval accuracy by 25% by leveraging LLM integration.
Dunlin AI
ML Intern
Improved financial prediction accuracy by 90%+ by designing and training two state-of-the-art ML models (DistilBERT and AutoGluon) for transaction analysis.
Education
Bachelor of Technology
Computer Technology
Yeshwantrao Chavan College of Engineering, Nagpur
Achievements
Winner
GPU-Accelerated Computing and Codeathon
Runner-up
Kaggle Datathon Competition