How aryangupta1211/agentic-trip-planner works
Project type and purpose
• This is a full-stack AI-driven travel planner.
• Frontend: Streamlit UI for user interaction.
• Backend: FastAPI service exposing REST endpoints.
• Core: A LangChain/LangGraph “agentic workflow” that orchestrates multiple sub-agents and real-time tools to build itineraries.
File structure
Top level:
• streamlit_app.py – entry point for Streamlit frontend
• main.py – entry point for FastAPI backend
• agent/ – defines the agentic_workflow (multi-agent orchestration)
• prompt_library/ – templates and prompt management
• tools/ – wrappers around each external data source (weather, currency, hotels, places, expense calc, arithmetic)
• utils/ – low-level API callers, config loader, model loader, save/output utilities
• config/ – configuration definitions (config.yaml)
• logger/, exception/ – cross-cutting concerns (logging, error handling)
• notebook/ – experimental prototyping
Extra clues from README
• Decoupled Streamlit ↔ FastAPI architecture
• Agentic workflow built on LangChain & LangGraph
• Two LLMs: Groq (fast) and OpenAI GPT-4 (reasoning)
• Real-time external services via tools (weather, hotel pricing, currency, places)
• Modular, Python-based, no database (stateless, all external data)
Diagramming conventions
• Use rectangles for internal modules (Streamlit, FastAPI, Agents, Tools, Utils).
• Blue for Presentation/API
• Green for Orchestration/Agents
• Orange for Tools/Utils
• Gray for cross-cutting concerns (config, logging, exceptions)
• User → Streamlit → FastAPI → Agent → Tools → External → back up the chain → Streamlit
Detailed capturing of essential elements
• Every directory’s responsibility: config, prompt_library, exception, logger, utils.
• How prompt_library feeds prompts into the agentic workflow.
• Model_loader’s role in connecting to LLMs.
• Demonstrate how save_to_document util outputs the final itinerary (e.g., PDF or markdown).