RMRowan maththiasinrowanmaththias.hashnode.dev·9h ago · 3 min readHow Tool Calling Works in AI AgentsLarge Language Models (LLMs) can understand requests, generate content, and reason through complex problems. However, an LLM by itself cannot directly interact with most external systems. Tool calling00
GJGeorge Jiehinmarginalium.hashnode.dev·1d ago · 4 min readWhat is Marginalium?I wanted a place to put my shower thoughts about AI. Explaining why I called it Marginalium requires a small detour through medieval manuscripts and a penis tree. Marginalium is an uncommon singular f00
RSRahul Sai Indeevar Vinrahul-ai.hashnode.dev·2d ago · 14 min readLLM Post-Training: From Full Fine-Tuning to PEFT, Adapters, and Soft PromptsLarge Language Models (LLMs) arrive with immense foundational knowledge acquired during pre-training. However, deploying them to specialized real-world domains—such as legal document parsing, medical 10
RSRahul Sai Indeevar Vinrahul-ai.hashnode.dev·2d ago · 12 min readDeep Dive into LLM Pre-Training: Paradigms, Mathematical Formulations, and Training ObjectivesBuilding effective Transformer-based Pretrained Language Models (T-PTLMs) requires understanding how models acquire general linguistic representations and adapt to specialized domains. Pre-training si10
RSRahul Sai Indeevar Vinrahul-ai.hashnode.dev·2d ago · 15 min readAdvanced Transformers: Positional Encodings (RoPE) & Efficient Attention Mechanisms (KV Caching, MQA, GQA, MLA)In modern Large Language Model (LLM) architectures, two foundational engineering bottlenecks dictate model scalability and inference latency: Representing Order: How self-attention mechanisms—which a12A
SASoftware Architecture Labinsoftwarearchitecturelab.hashnode.dev·Sep 2 · 6 min readDesigning Scalable AI Applications: Architecture Patterns Developers Should KnowIntroduction Building an AI-powered application is no longer only about connecting an application to an AI model and generating responses. The first generation of AI integrations focused on simple use00
DSDoogal Simpsonindoogal.dev·Aug 31 · 5 min readStop Terminal Prompting: A Markdown-First AI WorkflowQuick Answer: Relying on interactive terminal prompts for slow AI agents forces us into constant, exhausting context-switching. To reclaim your flow state, don't code in the terminal. Instead, draft y10
RSRahul Sai Indeevar Vinrahul-ai.hashnode.dev·Aug 28 · 9 min readUnderstanding LoRA: Parameter-Efficient Fine-Tuning for Modern LLMsLarge Language Models (LLMs) and foundation models like LLaMA, GPT, and ViTs have billions of parameters. As these models scale, traditional full fine-tuning—updating every weight in the network—becom10
ZZyVOPinzyvop.hashnode.dev·Aug 27 · 9 min readA 4B Open Model Matches GPT-5.6 Sol on Retrieval at 100× Lower CostWhat if one of the biggest problems with AI agents isn't that their models aren't powerful enough—but that we're using expensive general-purpose models for jobs that smaller models could learn to do? 00
DSDoogal Simpsonindoogal.dev·Aug 26 · 4 min readWhy You Should Wait 3 Months for Open-Source AIStop chasing the bleeding edge of proprietary AI models. By waiting just three months to adopt new technology, you allow cheaper, highly stable open-source alternatives to catch up in intelligence. Th00
