Authors: Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, Yuan Cao (2022)

This paper introduces the ReAct loop framework, which prompts LLMs to generate reasoning traces ("Thoughts") and task-specific actions ("Actions" e.g., tool calls) in an alternating manner. It demonstrates how reasoning improves tool accuracy and how tool results ground the model's reasoning.

Authors: Ehud Karpas, Omri Aschheim, Amit Sandwy, Barak Wanhon, Imri Almog, Leslie Valer, Yotam Shemesh, Asaf Dar, Yair Lichtig, Barak Lenz, Yuval Nov (2022)

Examines Modular Reasoning, Knowledge and Language (MRKL) systems. Formalized how neural networks (LLMs) can be integrated with symbolic utilities (calculators, databases, APIs) to solve math, logic, and factual recall tasks.

Authors: Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela (2020)

Introduced Retrieval-Augmented Generation (RAG). Shows how fine-tuning or prompting a model using documents retrieved by a dense retriever (like DPR) greatly reduces hallucinations on complex, open-domain questions.

Authors: Luyu Gao, Xueguang Ma, Jimmy Lin, Jamie Callan (2022)

Presents Hypothetical Document Embeddings (HyDE). Shows how having an LLM generate a hypothetical response, embedding that response, and using it to search a vector index yields superior zero-shot retrieval accuracy over standard queries.

Authors: Yu. A. Malkov, D. A. Yashunin (2016)

The foundational paper introducing Hierarchical Navigable Small World (HNSW) graphs. It shows how layering dense and sparse connections on a multi-layer graph enables sub-millisecond approximate nearest neighbour queries in high-dimensional vector spaces.

Authors: Noah Shinn, Federico Cassano, Beck Labash, Ashwin Gopinath, Karthik Narasimhan, Oren Etzioni (2023)

Introduces Reflexion, an architectural framework that equips agents with dynamic memory and evaluation loops to reflect on failure states, write self-critiques, and dynamically correct plans in future steps.

Author: Sebastian Raschka (Manning Publications, 2025)

An exceptional, step-by-step guide to coding your own GPT-like model in PyTorch. Covers data tokenization, building the multi-head attention network, pre-training, instruction fine-tuning, and evaluating LLM outputs.

Author: Chip Huyen (O'Reilly Media, 2022)

A widely respected guide for production machine learning systems. Covers data engineering, model selection, low-latency prediction pipelines, continuous monitoring, and ML feature stores.

Authors: Lewis Tunstall, Leandro von Werra, Thomas Wolf (O'Reilly Media, 2022)

Written by Hugging Face team members, this book is a deep guide to training, fine-tuning, and scaling transformers. Covers attention layers, translation, token classification, and accelerating inference.

Authors: James Phoenix, Mike Taylor (O'Reilly Media, 2024)

A practical handbook detailing structural prompt layout patterns, system prompts, few-shot tuning, output validation, and mitigation approaches for prompt injection vulnerabilities.