AI Season RAG — Code v2 (LangChain style)

Simpler teaching version modeled after rag-for-beginners:

• Flat numbered scripts (no utils/ package)
• LangChain loaders, splitters, Chroma, retrievers
• Local embeddings via HuggingFaceEmbeddings (free, no OpenAI key)
• Groq LLM via ChatGroq (not OpenAI)
• Mock mode when GROQ_API_KEY is missing

Session: How AI Answers From Your Data

---

Setup (Windows PowerShell)

cd rag-session-code-v2
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
Copy-Item .env.example .env

Optional Groq key in .env:

LLM_PROVIDER=groq
GROQ_API_KEY=gsk_your_key_here
GROQ_MODEL=llama-3.1-8b-instant

---

Scripts (run in order for first demo)

File	What it does
1_file_conversion.py	PDF/DOCX/TXT from data/raw/ -> docs/
2_ingestion_pipeline.py	Load, chunk, embed, store in db/chroma_db
3_retrieval_pipeline.py	Similarity search only
4_answer_generation.py	Retrieve + Groq (or mock) answer
5_recursive_chunking_demo.py	Compare chunking strategies
6_retrieval_methods.py	Similarity / threshold / MMR demos
7_full_rag_pipeline.py	End-to-end RAG with sources
8_failure_experiments.py	Out-of-scope, top_k, weak prompts
9_hybrid_retrieval.py	Vector + BM25 hybrid search
10_streamlit_app.py	Interactive AI Season assistant

Quick start

python 1_file_conversion.py
python 2_ingestion_pipeline.py
python 3_retrieval_pipeline.py
python 4_answer_generation.py
python 7_full_rag_pipeline.py
streamlit run 10_streamlit_app.py

---

vs rag-for-beginners

rag-for-beginners	This project
OpenAIEmbeddings	HuggingFaceEmbeddings (local)
ChatOpenAI	ChatGroq
docs/ tech company files	Refund policy + university FAQ
No mock mode	Mock fallback for teaching

vs rag-session-practical-code (v1)

v1	v2
utils/ modules + classes	Single-file scripts
Raw sentence-transformers + chromadb	LangChain wrappers
examples/slide_XX.py	1_...py numbering like rag-for-beginners

---

Troubleshooting

• No vector store: run 2_ingestion_pipeline.py first
• Slow first run: embedding model downloads once (~80 MB)
• Groq errors: set LLM_PROVIDER=mock in .env
• Re-index: delete db/chroma_db and run ingestion again