This repository contains a GenAI stack for Lahwita (Ontario LTB assistant), with working scaffolding for RAG, graph orchestration, and document processing. It is wired for:

• LangGraph orchestration
• Gemini API (text + embeddings)
• Pinecone for legal sources
• ChromaDB for CanLII case retrieval
• FastAPI backend + Streamlit client

Services (HTTP endpoints):

• GET /health: health check.
• POST /ai/chat: main form chat endpoint (discovery, Q&A, form fill, confirmation).
• POST /ai/file: PDF generation handoff (stubbed response).
• POST /lahwita/ai/indexr/: index configured document sources into Pinecone.
• POST /ai/consultant: consultant mode powered by CanLII case retrieval.

Facilities (platform capabilities):

• Dual RAG pipelines: Pinecone for LTB rules/forms and ChromaDB for CanLII cases.
• CanLII crawler + local PDF indexer for case ingestion.
• LangGraph graphs for chat, file generation, and consultant flows.
• Document processing utilities (PDF-to-text, chunking, preprocessing).
• Streamlit UI for Form + PDF and Consultant modes.
• Unit and integration tests in tests/.

User-facing modes (Streamlit UI):

• Form + PDF: multi-step form workflow (discovery → Q&A → fill → confirm) and PDF generation.
• Consultant: legal consultant chat using CanLII case retrieval and LTB rules.

Internal chat intents (used to pick prompts in /ai/chat):

• describing_situation: user describes their scenario (form discovery).
• asking_about_form: user asks questions about a known form (Q&A).
• answering_form_field: user provides field answers (form fill).
• other: fallback/confirmation.

The consultant pipeline is designed to use:

• LTB rules/instructions from Pinecone indexes.
• Official CanLII LTB cases from 2025 and 2026.

Case ingestion options:

• Crawl CanLII pages (default start URL is the 2026 LTB date index).
• Index local PDFs from data/cache/ltb_law_case without crawling.

The Streamlit consultant mode exposes a Refresh CanLII index before answering toggle to re-index before retrieval.

Key environment variables (see .env.example for the full list):

• Gemini: GEMINI_API_KEY, GEMINI_MODEL, GEMINI_EMBEDDING_MODEL
• Pinecone: PINECONE_API_KEY, PINECONE_HOST, PINECONE_CLOUD, PINECONE_REGION
• Indexes: INDEX_* for LTB rules, forms, and instructions
• ChromaDB: CHROMA_HOST, CHROMA_PORT, CHROMA_COLLECTION
• Document sources: DOC_* for ingestion paths/URLs

Build and run with Docker Compose:

docker compose up --build

The API container exposes port 8000 and runs:

uvicorn app.main:app --host 0.0.0.0 --port 8000

See STRUCTURE.md for the complete file map.