🤖 Gemini 2.5 Flash Conversational Chatbot

This project implements a multi-turn, stateful chatbot using FastAPI and LangGraph with Gemini 2.5 Flash for all NLU, Guardrail, and Knowledge Base tasks. It fulfills all the requirements of the Chatbot Development Assignment, including fuzzy time parsing, mid-conversation topic switching, and ethical guardrails.

⚙️ Technical Stack

• Framework: FastAPI (for the single /chat endpoint)
• Conversational Logic: LangGraph (State Machine / Agentic Flow)
• NLU/AI Model: Google Gemini 2.5 Flash
• Session State: In-memory checkpointer (MemorySaver from LangGraph)
• Deployment: Docker

🚀 Setup and Run

Prerequisites

1. Python 3.11+
2. Docker (highly recommended)

1. Local Setup

1. Navigate to project folder.
2. Install dependencies:

   pip install -r requirements.txt

3. Set API Key: Create a file named .env in the project root and add your key:

   GOOGLE_API_KEY="YOUR_API_KEY_HERE"
   

4. Run the server:

   uvicorn app.main:app --reload

2. Docker Setup (Recommended)

1. Build the Docker image:

   docker build -t chatbot-gemini .

2. Run the container:

   # Pass the API key as an environment variable
   docker run -d -p 8000:8000 --env GOOGLE_API_KEY="YOUR_API_KEY_HERE" chatbot-gemini

   The bot will be accessible at http://localhost:8000.

🧪 Testing the Chatbot (Full User Journey)

The bot maintains session state via the user_id passed in the JSON payload.

Endpoint: POST http://127.0.0.1:8000/chat Headers: Content-Type: application/json

Use a tool like curl to test the full conversational flow:

Step	User Input (message field)	Expected Bot Action
1 (Fuzzy Time)	I want to book a table this weekend or maybe Monday morning.	Clarifies ambiguous dates.
2 (Clarification)	Sunday, please.	Sets the date; asks for party size.
3 (Party Size)	For four people.	Sets size; asks for final confirmation.
4 (Topic Switch)	By the way, what's the capital of Australia?	Answers the factual question, then resumes the confirmation question (e.g., "Now, back to your reservation... Should I confirm that?").
5 (Guardrail)	You're an idiot.	Replies with: "Let's keep our conversation respectful, please."
6 (Final Confirm)	Yes-please confirm the reservation.	Confirms the reservation details.

---

This completes all phases of the development assignment. The codebase is now structured, functional, and ready for review and deployment.

Do you have any final questions or require any specific explanation of the code structure before concluding?

🐳 Run, build and run with Docker Compose

Here are the exact commands you requested for running manually, building the Docker image, and running with Docker Compose.

1. Run locally (module entrypoint, works even if the uvicorn console script is not on PATH):

python -m uvicorn main:app --reload

2. Build the Docker image (from the repository root):

docker build -t chatbot-memory-test .

3. Bring up the service using Docker Compose (detached):

docker compose up -d

Notes:

• If you want docker compose to rebuild the image before starting, run docker compose up -d --build.
• If you prefer the uvicorn CLI command (instead of python -m uvicorn), ensure the Python scripts location (e.g. ~/.local/bin) is on your PATH or install into an activated virtualenv.