Exploring the Frontier
of AI & Machine Learning.
Deep dives into Neural Networks, LLMs, and the future of technology. Join me as I decode the algorithms shaping our world.
Latest Writing
Thoughts on AI, ML, and software engineering
When vector embeddings aren’t enough: A lesson in precision, pricing, and agentic workflows.
Large language models, like the brilliant minds behind GPT-3, GPT-4, and LLaMA, are learning to think and write just like us humans! By Training huge amounts of text, they're getting smarter and smarter, understanding the context of conversations, answering questions, finishing stories, and even creating their own! Take LLaMA, for example, a family of models created by Meta. It's like a super-smart language assistant, efficient and powerful, excelling at tasks like translating languages, summarizing long texts, and answering your questions. But as these models grow bigger and smarter, they need a lot of energy and computing power to learn and work
Selected Projects
Open source work and experiments
Developed a citation-aware AI agent for government services, delivering answers with direct document sourcing to ensure transparency Built an admin control plane allowing dynamic document management and configurable chunking strategies to optimize retrieval accuracy Integrated a real-time observability dashboard to track active sessions, token consumption, and operational costs
AgentBox is a foundational security layer designed to operate between an AI agent's code and the host operating system. It intercepts potentially hazardous operations-such as file system access, network connections, or command execution-and evaluates them against a user-defined security policy.
An intelligent invoice extraction system that combines Google Gemini OCR with OpenAI GPT for accurate, structured data extraction from PDF invoices.