みなさん、こんにちは
Minasan, kon’nichiwa
Hi, I’m Ajay Banstola; a software engineer and AI enthusiast with over 4 years of industry experience and a deep curiosity for building intelligent systems. I’m currently pursuing my Master’s in Computer Science at the University of Louisiana at Lafayette, where I also contribute to research on AI-driven program analysis and fuzzing techniques. My work spans from building scalable backend systems and cloud-native applications to deploying machine learning models in real-world environments. I enjoy working at the intersection of software engineering and AI, whether that means integrating large language models into production workflows or designing systems that make data-driven decisions. I recently joined Blue Partner as an AI Integration Engineer Intern last summer, and I got to learn a lot creating meaningful tech that blend smart automation with human impact. I also got to keep myself up to date with the latest technology.
Outside of tech, I love staying active and competitive whether it’s playing badminton, table tennis, or a strategic game of chess. I also unwind by diving into the world of manga, which has been a big part of my life and even sparked my interest in learning Japanese. It’s been a fun challenge picking up the language, and it adds a whole new layer of appreciation to the stories I enjoy. These hobbies keep me balanced and curious, both in and out of work.
LinkedIn
GitHub
From Simple to Smart: Understanding LangChain and AutoGen Artificial Intelligence is evolving faster than ever; and with it, new tools are emerging that make it easier to build complex AI systems. Two of the most powerful ones today are LangChain and AutoGen.
In this post, we’ll start simple; understanding what they are, why they matter; and then dig deeper into their core architecture and advanced usage patterns.
1. The Basics: What Problem Do They Solve? Let’s say you have a large language model (LLM) like GPT-4. It’s great at generating text, but it’s just one piece of the puzzle.
...
Welcome! In this post, we explore two key concepts for building intelligent, context-aware AI systems: Retrieval-Augmented Generation (RAG) and Model Context Protocol (MCP).
These frameworks tackle a critical challenge: how to make LLMs more accurate, grounded, and capable of complex interactions.
RAG: Retrieval-Augmented Generation Purpose: To empower language models with access to external knowledge at inference time.
How It Works: Retriever module fetches relevant documents from a vector store or indexed database based on the user query. Generator (usually an LLM) uses both the query and retrieved documents to generate an informed response. Why It Matters: Reduces hallucination. Enables real-time factual accuracy. Supports plug-and-play updates without retraining the model. Example Use Cases: Chatbots that answer based on internal company wikis. Research assistants grounded in academic papers. Domain-specific assistants (e.g., legal, medical). MCP: Model Context Protocol Purpose: To manage multi-turn memory, task orchestration, and context flow in multi-agent or tool-augmented AI systems.
...
When diving into the world of AI, two classes of models dominate the conversation: Large Language Models (LLMs) and Diffusion Models. While both are powerful generative AI technologies, they operate on fundamentally different principles and serve distinct purposes.
This post explores what diffusion models are, how they work, and how they fundamentally differ from LLMs.
Diffusion Models: The Generative Artists Purpose: To generate high-quality images, audio, and other continuous data by learning to reverse a gradual noise-adding process.
...