src/README.md

Handbook Generator (AI Engineering Assignment)

Same technology as the policy project: Google ADK, LiteLLM (OpenAI), and Streamlit. The UI talks to the ADK agent directly (no API). The agent uses a RAG tool (ChromaDB) to answer from uploaded PDFs.

Features

• PDF upload — Upload PDFs; text is extracted, chunked, embedded (OpenAI), stored in ChromaDB (local).
• Chat — ADK agent runs in the same process; agent calls RAG tool query_uploaded_documents and answers with the LLM (OpenAI via LiteLLM).
• Handbook generation — Request a 20,000+ word handbook; generation uses RAG and runs section-by-section.
• Export — Download the handbook as Markdown.

Architecture

• Streamlit (streamlit_app.py) → UI; imports runner_app to run the agent directly.
• runner_app.py → ADK Runner + session; run_chat(message) runs the agent (sync wrapper around runner.run_async).
• agent.py → ADK Agent (LiteLLM/OpenAI), tools = [query_uploaded_documents].
• RAG (rag.py + rag_tools.py) → ChromaDB + OpenAI embeddings.

Setup

1. Python

Use Python 3.10+.

2. Install dependencies

cd ass2
pip install -r requirements.txt

3. Environment

Create a .env file in ass2 (see .env.example):

OPENAI_API_KEY=sk-your-openai-api-key-here
MODEL=gpt-4o

Run (single command)

cd ass2
streamlit run streamlit_app.py

Open http://localhost:8501. No separate API server.

How to use

1. Upload PDFs — In "Upload PDFs", select PDFs and click Index PDFs.
2. Chat — In "Chat", ask questions; the ADK agent uses the RAG tool and answers from your documents.
3. Generate handbook — In "Generate Handbook", enter a topic and click Generate handbook, then download as Markdown.

Project structure (all in ass2)

File	Purpose
streamlit_app.py	Streamlit UI (upload, chat, handbook)
runner_app.py	ADK Runner + session; run_chat(message) for Streamlit
agent.py	ADK agent (LiteLLM/OpenAI) + RAG tool
prompt.py	Agent name, description, instruction
rag_tools.py	ADK tool: query_uploaded_documents
callback.py	ADK callbacks
rag.py	ChromaDB + OpenAI embeddings
pdf_processor.py	PDF text extraction and chunking
handbook_generator.py	20k-word handbook generation
config.py	Settings and paths

Tech stack

• Agent: Google ADK, LiteLLM (OpenAI)
• RAG: OpenAI embeddings, ChromaDB (local)
• UI: Streamlit (agent runs in-process, no FastAPI)