How kaushalprajapatikp/agentic-ai-webapp-with-multiple-ai-agents-and-deployment works
Project type and purpose
• This is a full-stack, end-to-end “Agentic AI” application built on Streamlit.
• Purpose: Expose an AI “agent” pipeline (graph of nodes and tools) via a web UI (app.py), where user inputs travel planning queries and the system orchestrates LLM calls, state management, tool execution (search, synthesizer, human review), and vectorstore memory.
File structure
Top-level and key directories:
• app.py + README.md + Dockerfile: entrypoint, documentation, containerization
• src/AgenticAI/main.py: core orchestration script
• src/AgenticAI/Graph: graph_builder.py – constructs the agent’s execution DAG
• src/AgenticAI/Node: individual processing units (basic_chatbot_node, travel_agent, etc.)
• src/AgenticAI/LLM: groqllm wrapper – abstracts calls to the LLM service
• src/AgenticAI/Tool: external tools & submodules (search_tool, travel orchestrator/evaluator/workers/human_review/synthesizer)
• src/AgenticAI/State: state.py and travel_state.py – holds conversation and travel planning context
• src/AgenticAI/VectorStore: embedding storage & retrieval
• src/AgenticAI/UserInterface/StreamlitUI: Streamlit pages and layout logic
Insights from README
• Confirms: built with Streamlit (app_file: app.py), Apache 2.0 license.
• Short description: “Refined langgraphAgenticAI,” i.e., a language-graph agent.
Main components
a. Frontend (User Interaction)
• Streamlit UI (app.py + StreamlitUI folder)
b. Agent Orchestrator
• main.py → uses GraphBuilder to assemble a Graph of Nodes
c. Nodes (Compute Units)
• Basic/Chatbot nodes, TravelAgent node
d. State Manager
• state.py, travel_state.py – shared context across nodes
e. LLM Service
• groqllm wrapper calls external LLM API (e.g., Groq)
f. Tooling Subsystem
• search_tool (web search API), travel tools (orchestrator, evaluator, synthesizer, human_review, workers)
g. VectorStore
• Embedding store for retrieval, memory, caching
Relationships and interactions
• User submits input via Streamlit → forwarded to main.py
• main.py invokes GraphBuilder → returns Graph of Node objects
• Graph execution iterates Nodes in sequence/branches:
• Each Node reads/writes to State Manager
• Nodes call LLM wrapper or Tooling functions as needed
• Tooling subsystem may call external APIs or human-in-the-loop service
• VectorStore accessed by nodes for memory/embeddings
• Final result sent back to UI for display
Architectural patterns and design principles
• Modular “Graph of Nodes” pattern: graph_builder configures flow, supports extension
• Separation of concerns: UI / Orchestrator / Nodes / LLM / Tools / State / Storage
• Dependency injection via GraphBuilder
• External service wrappers (LLM, search) abstracted behind interfaces
• Stateful vs. stateless components clearly separated
Technologies and frameworks
• Python 3.x, Streamlit
• groqllm (LLM client)
• Potential HTTP clients for search and tool APIs
• Docker for containerization
• VSCode/GitHub Actions (main.yml) for CI/CD