Author's Own Writing

Peter's Weblogs

Essays, technical deep-dive guides, and reading directories written and curated by Pethuru Raj — spanning kernel methods, graph neural networks, RAG architectures, LLM training, and enterprise AI system design. Click any card to open it.

15 resources

Essays & Explainers

Plain-English deep dives
How Modern AI Clusters Load Balance GPU Inference for LLMsEssay
Routing an LLM request is not like routing an HTTP request. This is a technical walkthrough of the mechanisms production clusters use to keep GPUs saturated — from KV-cache-aware routing to disaggregated prefill/decode serving.
www.peterindia.net/HowModernAIClustersLoadBalanceGPUInference.html
Legacy Application Modernization: From COBOL to Cloud-Native & Agentic AIEssay
A working COBOL core banking system as the running example: the proven "7 R's" of modernization, and the 2026 shift toward agentic AI-driven transformation and AI-agent orchestration layers.
www.peterindia.net/LegacyApplicationModernizationMechanisms.html
CPU, GPU, TPU, NPU, LPU, DPU & QPU ExplainedEssay
Seven processor types, seven different bottlenecks. A practical guide to what each chip is actually built for, how they work together in a modern AI stack
www.peterindia.net/ProcessorTypesExplained.html
LLM Training and Inferencing Optimization TechniquesEssay
Training and serving a large language model are two different engineering problems that happen to share a GPU. Here's what actually determines cost and speed on each side — and why the techniques that help one side often don't help the other.
www.peterindia.net/LLMTrainingAndInferencingOptimizationTechniques.html
Graph Neural Networks: Learning From Connected DataEssay
Most machine learning models work with structured tables, images, or text. But many real-world problems involve relationships and connections
www.peterindia.net/GNNs.html
Steps to Build Powerful LLMsEssay
Understand exactly where web scraping, tokenization, transformer architectures, causal language modeling, RLHF, gradient clipping, adaptive learning, and supervised fine-tuning each fit into the LLM pipeline
www.peterindia.net/LLMBuildingSteps.html
Understanding Graph Neural NetworksEssay
A practitioner's blog on GNNs — message passing, GCN/GAT/GraphSAGE/GIN, real applications, and where the field is heading in 2026.
www.peterindia.net/UnderstandingGNNs.html
Enterprise AI System Design ConceptsEssay
An AI system can have a powerful model and still collapse under production traffic. Once AI moves beyond a demo, model quality is only one part of the architecture
www.peterindia.net/EnterpriseAISystemDesignConcepts.html
The Kernel TrickEssay
An inspiring, plain-English tour of kernel methods in machine learning — what they are, why they exist, and why SVMs still rely on them.
www.peterindia.net/KernelFunctions.html

Technical Deep-Dive Guides

Structured reference articles
AI Model ArchitecturesGuide
A technical guide comparing GPT, MoE, LRM, SLM, and LAM architectures, and how to choose the right model for the right task.
www.peterindia.net/LanguageModels.html
Graph Neural NetworksGuide
How GNNs learn from nodes and edges via message passing, the four core architectures, real applications, and 2026 trends.
www.peterindia.net/GNNs.html
Trends and Transitions in RAGGuide
How Retrieval-Augmented Generation evolved from a simple pipeline into agentic, multimodal, real-time retrieval architectures in 2026.
www.peterindia.net/RAGTypes.html
Enterprise AI System DesignGuide
Twelve distributed-systems concepts — caching, circuit breakers, rate limiting, and more — that keep RAG and agentic AI reliable in production.
www.peterindia.net/EnterpriseAISystemDesignConcepts.html
Steps to Build Powerful LLMsGuide
The 6-step pipeline for building large language models — data collection, tokenization, pretraining, RLHF alignment, deployment, and evaluation.
www.peterindia.net/LLMBuildingSteps.html

Directories

Curated link collections
Weblogs DirectoryDirectory
A hand-checked directory of tech weblogs covering Java & XML, software development, databases, security, wireless, and EAI/SOA topics.
www.peterindia.net/Weblogs.html

No matches. Try a different keyword.