Gen-AI Developer Classroom notes 12/Feb/2026

Ensuring only updated docs are indexed

• use the following code

directory_loader = DirectoryLoader(
    path="../data/updates/IT_Helpdesk_KB_Articles_v2",
    glob="*.txt",
    loader_cls=TextLoader,
)
documents = directory_loader.load()
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=100,
    chunk_overlap=20,
)
chunks = text_splitter.split_documents(documents)
embedding = VertexAIEmbeddings(
    model_name="text-embedding-005")
vector_store = Chroma(
    collection_name="kb_collection",
    embedding_function=embedding,
    persist_directory="../vectordb/kb_collection_db_sample1",
)
# only changed docs and reindex
result = index(
    docs_source=documents,
    record_manager=sql_record_manager,
    vector_store=vector_store,
    cleanup='incremental',
    source_id_key='source'
)

• Refer Here for updating indexes.

Project for RAG: – HR Helpdesk

• The core idea of this project is we would be having HR documents (knowledge bases)
• For the purposes of this project i would have this docs in folder, in your organization they might be on confluence, wiki pages
• What we would be building:
  • Chatgpt kind of interface
  • User asks the question and RAG should respond
  • This is enterprise RAG, so it expected to be grounded
  • We should prove the precision, fairness of RAG before deploying
• Deployment possibilities
  • Models:
    • Cloud:
      • AWS
      • Azure
      • GCP
    • On-prem:
      • ollama
  • RAG & Vector db
    • Deployment options
      • on-prem
      • cloud
• Refer Here for the git repo.

Steps:

1. synthetic data creation:
2. understanding data
3. indexing
4. vector stores
5. Prompts
6. Retrieval
7. Grounding
8. Scores
9. Deployment