LLM Hub — Models, Inference, RAG, LLMOps & Gateways

This hub is the starting point for the LLM material on peterindia.net. Each spoke below is a directory or technical guide of its own, grouped by the stage of the LLM lifecycle it covers.

30Spoke pages

7Topic areas

Models, Assistants & Architectures

Which language models exist, how the architectures differ, and where to find open, compact and assistant models.

01Language Models: Architectures and TypesGPT, MoE, reasoning, small and action models explained, plus a directory of every model family from LLMs to LQMs. 02Large Language Models (LLMs)Directory of 43 frontier, open-weight and earlier-generation models and assistants. 03Open-Source LLMsLeading open-weight language models you can download, run and adapt. 04Small Language Models (SLMs)Compact models built for on-device and low-cost use, with how they compare. 05AI AssistantsChatbots and AI search, writing tools, coding assistants and agents, and conversational AI platforms: 62 entries to compare. 06Vision & Multimodal Language Models (VLMs, MLMs)Models that read images, charts and documents and handle audio and video, from proprietary APIs to open and on-device models. 07Multimodal Language Models for OCRVision-language models and OCR engines for document AI, such as DeepSeek OCR, PaddleOCR and Mistral OCR. 08World Action Models (WAMs)How joint video-action models let robots imagine a future state and the motion needed to reach it. 09NLP Models for Sentiment AnalysisLanguage models and NLP approaches for classifying sentiment in text. 10Generative AI Platforms and ToolsImage, video and audio generation tools, developer platforms and model hubs, and industry applications. 11Large Reasoning Models (LRMs)Models that think before answering, with open-weight options, training methods and benchmarks. 12Diffusion Language Models (dLLMs)Models that generate text by refining a whole sequence in parallel, from Mercury to open LLaDA and Dream. 13Large Action Models (LAMs)Computer-use, browser and GUI agents that turn instructions into actions. 14Large Quantitative Models (LQMs)Models grounded in numbers and science, from SandboxAQ to time-series foundation models. 15Decision ModelsJev and open-source alternatives for typed, auditable AI decisions.

Building, Training & Learning

How models get built, plus the libraries and learning resources for developers working with them.

16Steps to Build an LLMThe training pipeline from data collection and tokenization to alignment and evaluation. 17LLM Training & Inference OptimizationParallelism, memory and fault tolerance for training; ten inference mechanics and the production levers that cut serving cost. 18Python Libraries for LLM ApplicationsTen libraries for model access, orchestration, RAG, fine-tuning and evaluation. 19AI & LLM Learning ResourcesFree courses, guides and repositories for learning LLMs, agents and prompt engineering. 20Datasets for Generative AI47 open datasets and resources for pre-training, fine-tuning and evaluating generative AI models: text, instruction, code, multimodal and benchmarks.

Inference & Serving

Running models efficiently, on a laptop or across a GPU fleet.

21LLM Inference Platforms and ToolsLLM API providers, GPU and fast-chip clouds, routers, serving frameworks and local apps, plus vLLM, SGLang, TensorRT-LLM, llama.cpp and Ollama compared. 22LLM Model Formats ExplainedHow GGUF, GPTQ, AWQ and EXL2 differ and when to use each. 23The Production AI Inference Stack and GPU Load BalancingvLLM, llm-d and distributed serving, prefill/decode disaggregation, and the routing mechanisms that keep GPUs saturated.

RAG & Retrieval

Grounding models in your own data with retrieval-augmented generation.

24RAG Architectures, Chunking and RerankingThe three-part system, evolution to agentic RAG, ten chunking strategies, five reranking models, failure modes, metrics and a production checklist.

LLMOps, Observability & Gateways

Operating LLM applications in production: tooling, tracing and the traffic control layer.

25LLMOps, Observability and ToolsOperating LLM applications: a nine-stage lifecycle, a checklist and risks, plus a directory of 47 LLMOps and observability tools. 26AI & LLM Gateways30 gateway platforms for routing, governing and securing LLM, MCP and agent traffic, plus the 12 capabilities that define an AI gateway.

Hallucination & Output Quality

Keeping model output accurate and grounded.

27LLM Hallucination MitigationFour mechanism classes and eight techniques in depth, from RAG and reranking to confidence gates.

Assistants & Applications

Products built on language models: chatbots, writing tools, coding assistants and conversational AI platforms.

28AI Chatbot PlatformsDirectory of 30 AI-enabled chatbot and conversational AI platforms for lead generation, customer support, sales and engagement. 29AI Chatbot Development PlatformsPlatforms and frameworks for building conversational AI bots, compared on features, pricing and capabilities. 30AI Coding AgentsAgentic coding tools that plan, write, test and review code across a repository.

Frequently Asked Questions

What is an LLM hub?

This LLM hub is an index of the guides and directories on peterindia.net that cover large language models: the models themselves, building and training them, inference and serving, retrieval-augmented generation, LLMOps and observability, AI gateways, hallucination control, and assistants and chatbots.

Where should I start if I am new to large language models?

Start with Language Models: Architectures and Types and Large Language Models for the landscape, then Steps to Build an LLM for how models are trained. If you want to build applications, continue with the Python libraries, RAG architectures and LLMOps guides.

What is the difference between an open-source LLM and a small language model?

An open-source or open-weight LLM publishes its model weights so you can download, run and adapt it. A small language model is defined by size: it has far fewer parameters so it can run on laptops, phones and edge devices at low cost. Many small models are also open-weight.

What is an AI gateway and why do LLM applications need one?

An AI or LLM gateway sits between applications and model providers. It routes requests, enforces access and spending limits, applies guardrails, caches responses and records observability data. It increasingly also governs MCP tool calls and agent-to-agent traffic.

How can hallucinations in LLM output be reduced?

Common approaches include grounding the model in retrieved data with RAG, constraining outputs with schemas and guardrails, using stronger or reasoning models for hard tasks, and evaluating outputs automatically and with human review. The hallucination mitigation guide covers four mechanism classes and eight techniques.