🦃 Training Demos — Copilotclaw

A public repo of AI/ML training notebooks managed by Crunch (Copilotclaw's AI agent).

Notebooks are designed to run in Google Colab with one click — no local install required.
🦙 Ollama variants run fully offline on your own machine.

---

📚 Notebooks

Notebook	Topic	Open	Explainer
human-rag-exercise.ipynb	Human RAG — understand retrieval by doing it on paper first		(start here!)
simple-rag-demo.ipynb	RAG with FAISS + SentenceTransformers
simple-rag-demo-ollama.ipynb	🦙 Local RAG with Ollama — same pipeline but with real LLM generation, fully offline
qwen35-tiny-rag-demo.ipynb	👁️ Qwen 3.5 Tiny + Vision RAG — 0.8B–9B models with native image understanding, fully offline or via DashScope API		Ollama or ALIKEY
simple-rag-demo-java.ipynb	☕ RAG in Java — same pipeline in pure Java with TF-IDF (no Python libs, runs in Colab via javac/java)		(compare with Python!)
ragmcp-start.ipynb	🚀 RAG MCP Starter — clone the full ragmcp repo and explore all its notebooks from Colab		(start here for ragmcp!)
bm25-explainer.ipynb	🔍 BM25 vs Embeddings — why keyword search still wins for product data; hands-on BM25 from scratch + live comparison		(no API key needed!)
rag-techniques-ollama.ipynb	🔬 Advanced RAG Techniques — HyDE, MultiQuery, Step-Back, Re-Ranking, BM25, Dense Embedding with Ollama
ragas-evaluation-ollama.ipynb	📏 RAGAS Evaluation — score your RAG pipeline on Faithfulness, Answer Relevancy, Context Precision & Recall using Ollama — no API key needed		(after rag-techniques!)
ragas-evaluation-guide.ipynb	📊 RAGAS Evaluation Guide — comprehensive walkthrough of all 5 RAGAS metrics (OpenAI + Ollama), diagnostic table, multi-question examples		(deep dive into RAGAS!)
ragas-grok-tutorial.ipynb	🦾 RAGAS + Grok Tutorial — evaluate your RAG pipeline with the free Grok API (xAI); all 5 metrics, climate-science sample data, per-question score breakdown, good vs bad answer analysis		(free Grok key required)
ragas-ali-tutorial.ipynb	🐉 RAGAS + Alibaba Qwen Tutorial — same 5-metric RAGAS evaluation using the Alibaba DashScope Qwen API (OpenAI-compatible); generous free tier, no second key for embeddings		(DashScope key required)
ragmcp/GraphRAG/Historical_GraphRAG_Demo - Copy.ipynb	🕸️ Historical GraphRAG — build a knowledge graph from historical texts using LLMs + Ollama, fully self-contained
graphrag-neo4j-demo.ipynb	🗄️ GraphRAG with Neo4j — install Neo4j Community in Colab, build a knowledge graph, query it with LangChain Neo4jGraph + GraphCypherQAChain; includes vector index + semantic search (Groq free tier LLM)		(free Groq key optional)
agentic-rag-demo.ipynb	🤖 Agentic RAG with Grok + LangGraph — self-correcting pipeline: Retrieve → Generate → Evaluate → loop until verified; uses Grok structured output for hallucination detection		(xAI API key required)
autogen-agentic-rag-demo.ipynb	🏢 Agentic RAG with Microsoft AutoGen + Grok — same Evaluator-Optimizer loop but using AutoGen's conversational FSM: agents literally talk to each other; a Python speaker-selection function replaces graph edges		(compare with LangGraph!)
Teaching_Agentic_Retrieval_in_Colab.ipynb	🧑‍🏫 Teaching Agentic Retrieval in Colab — step-by-step introduction to agentic retrieval pipelines, designed for teaching/classroom use		(classroom-ready!)

---

🔍 What's in the RAG notebook?

The simple-rag-demo.ipynb notebook walks through the full RAG pipeline:

1. Build a knowledge base — a small set of documents about RAG, FAISS, and local LLMs
2. Embed documents using sentence-transformers (all-MiniLM-L6-v2, ~80MB, runs in Colab free tier)
3. Index with FAISS — efficient vector similarity search by Meta
4. Query the index — retrieve top-k most relevant documents for any question
5. Build a prompt — see how retrieved context becomes an LLM input
6. Exercises — extend it yourself

No API keys. No cloud credits. Runs fully in Colab.

---

🚀 Quick Start

Colab (recommended): Click the badge above → runs in your browser in ~60 seconds.

Local:

git clone https://github.com/Copilotclaw/trainingdemo
cd trainingdemo
pip install sentence-transformers faiss-cpu jupyter
jupyter notebook simple-rag-demo.ipynb

---

🏫 PIRAGE RAG Trainingskurs (Deutsch)

Ein vollständiger 3-Tage RAG-Kurs auf Deutsch — basierend auf dem LearnRAG-Material. Kein API-Key, kein lokales Setup — läuft in Google Colab.

PIRAGE: Parse → Index → Retrieval → Augmented Generation → Evaluation

Session	Titel	Notebook	Colab
1	Grundlagen von RAG + PIRAGE-Framework	01_grundlagen_rag.ipynb
2+3	Naive RAG — Einrichtung & Chunking	02_naive_rag.ipynb
4–12	Hybrid RAG, GraphRAG, Agentic RAG…	(coming soon)	—

→ Vollständige Kursübersicht

---

🛠️ Generating Explainers

Each notebook has an auto-generated HTML explainer. To regenerate or add one for a new notebook:

# Install the one dependency
pip install openai

# Generate (requires ALIKEY or AZURE_API_KEY env var)
python scripts/generate_explainer.py my-notebook.ipynb

# Test without an LLM key
python scripts/generate_explainer.py my-notebook.ipynb --no-llm

# Custom output path
python scripts/generate_explainer.py my-notebook.ipynb --output my-notebook-explainer.html

The script calls Ali qwen3-coder-plus (or Azure as fallback) to write the explainer content, then renders it as a polished HTML page with hero, steps, key concepts, and a CTA.

Auto-generation: add .github/workflows/generate-explainers.yml (file is ready in the repo root — needs a user with workflows permission to commit it) to auto-regenerate on every notebook push.

---

🦃 About

This repo is managed by Crunch — an AI agent running on GitHub Actions. Notebooks are generated and committed autonomously based on training needs.

More notebooks coming as training topics expand.