Research & Science
RAG.md · 17 documents
Effect-Atom MCP Enhancement Research Report
Analyzes an Effect-based MCP server's architecture and proposes enhancements for retrieval, validation, and atom-react integration.
Deep Learning for AI
Summarises a survey paper on deep learning origins, breakthroughs, and future challenges for AI researchers.
AI Scientist Lab Notebook — Architecture (V1)
Defines the full architecture for an AI scientist lab notebook with RAG, ingestion, and plotting pipelines.
Technical Book Rack RAG — Implementation Plan
Plans a 10-phase fork of a PDF RAG system into a technical-book RAG with concept-aware chunking and book metadata.
四、随机化 SVD
Teaches randomized SVD through three concrete applications: word vector projection, video background removal, and topic modeling.
draft
Lists and briefly describes transformer-based computer vision papers, from standard architectures to few-shot learning, with diagrams and code links.
The Annotated Encoder-Decoder with Attention
Implements an encoder-decoder with Bahdanau attention using GRUs, closely following the 2015 neural machine translation paper.
RFC-001: Continuity
Proposes replacing a multi-runtime memory system with a single Go binary that captures, categorizes, and retrieves session memories for Claude Code.
📚 RAG with Mistral - Technical Documentation
Documents a RAG system that retrieves local document context and generates answers using Mistral via Ollama.
llm-hallucination-survey
Curates a structured bibliography of 100+ papers on LLM hallucination, organised by evaluation, source, and mitigation.
NLP - Dialogue System
Lists 12 influential papers on multi-turn dialogue systems with conference venues and one-sentence summaries of each contribution.
RALM_Survey
Catalogs over 100 papers on retrieval-augmented language models, organized by definition, retriever, LM, enhancement, data source, and application.
REFRAG: Retrieval-Enhanced Fine-Grained Retrieval Augmented Generation
Explains REFRAG, a RAG variant that passes retrieved vectors directly to an LLM to skip re-encoding, and provides a DSPy-based Python implementation.
Notes
Collects RAG optimization techniques, a MicroSaaS business plan, and a LangChain development strategy into a single personal reference note.
RAG-LLaMA3 AI Project
Describes a RAG pipeline using LLaMA3 to answer questions about DNALinux bioinformatics tools, with code to build your own.
Why We Move from RLM to RLM-Graph
Explains why a team replaced unstructured-text RLM with a graph-based environment to make recursive exploration reliable and auditable.
THE LJPW CODEX
Defines a semantic-first framework with four principles (Love, Justice, Power, Wisdom) and a Python simulation engine for modeling meaning dynamics.