SitBack — AI-Powered Freight Negotiation Platform

Autonomous AI agents that negotiate freight rates with multiple transporters in parallel — so you don't have to.

Live Demo → v0-negotiation-agent-ui.vercel.app

---

What is SitBack?

SitBack is a B2B logistics SaaS platform that automates freight rate procurement for manufacturers and shippers. Instead of manually calling transporters and negotiating rates, clients submit a shipment request and an AI agent handles the entire negotiation — running parallel conversations with multiple LSPs (Logistics Service Providers), countering quotes, and recommending the best deal.

The core loop:

Client posts RFQ → AI agent negotiates with N transporters in parallel
→ Scores each LSP → Recommends best deal → Client accepts / pushes another round

---

Key Features

• Parallel AI Negotiation — the agent simultaneously negotiates with multiple transporters, not sequentially
• LangGraph-powered agent — stateful multi-step reasoning using Google Gemini as the LLM backbone
• Live WebSocket updates — negotiation state pushed to the frontend in real time
• Transporter Portal — LSPs can log in, view open RFQs, and submit bids directly
• Scoring & Recommendation — each LSP is scored on Quote, Profile, and Reliability; the agent explains its recommendation in plain language
• Simulator — a transporter simulator allows testing the negotiation loop without real LSPs

---

Tech Stack

Layer	Technology
Backend Framework	FastAPI
ASGI Server	Uvicorn
AI Agent Framework	LangGraph
LLM	Google Gemini (langchain-google-genai)
Real-time	WebSockets
Data Validation	Pydantic v2
Data Store	JSON flat files (/data folder)
Frontend	Vercel (separate repo)

---

Project Structure

LN-Agentic-Hackathon/
├── main.py                  # FastAPI app entry point, route registration
├── websocket_manager.py     # WebSocket connection & broadcast manager
├── requirements.txt
├── .env                     # Environment variables (not committed)
│
├── agent/                   # LangGraph agent definition & negotiation graph
├── llm/                     # Gemini LLM client setup & prompt templates
├── logic/                   # Core negotiation logic, scoring, recommendation
├── simulators/              # Transporter simulator for testing negotiations
├── routers/                 # FastAPI route handlers (RFQ, transporter, negotiation)
├── models/                  # Pydantic models / schemas
├── db/                      # Data access helpers (reads/writes JSON files)
└── data/                    # Flat-file data store (JSON objects and arrays)

---

Getting Started

Prerequisites

• Python 3.10+
• A Google AI Studio API key (Gemini)

1. Clone the repo

git clone https://github.com/dodomyg/LN-Agentic-Hackathon.git
cd LN-Agentic-Hackathon

2. Create and activate a virtual environment

python -m venv venv
source venv/bin/activate        # macOS / Linux
venv\Scripts\activate           # Windows

3. Install dependencies

pip install -r requirements.txt

4. Set up environment variables

Create a .env file in the root directory:

GEMINI_API_KEY=your_google_gemini_api_key_here

5. Run the server

uvicorn main:app --reload

The API will be available at http://localhost:8000.
Interactive docs at http://localhost:8000/docs.

---

How the Negotiation Works

1. Client submits an RFQ with origin, destination, truck type, capacity, budget, and target transporters.
2. The LangGraph agent initialises a negotiation session and notifies all selected LSPs.
3. Each LSP submits an initial quote (either via the Transporter Portal or the simulator).
4. The agent evaluates each quote against the budget and benchmark, then decides to accept, counter, or drop each LSP independently.
5. Counter-offers are sent and the cycle repeats for N rounds.
6. Once stopping conditions are met, the agent scores all LSPs and surfaces a recommendation with reasoning.
7. The client accepts, rejects, or pushes one more round from the Recommendation Panel.
8. All updates are streamed live to the frontend via WebSocket throughout.

---

API Overview

Method	Endpoint	Description
POST	/rfq	Create a new shipment RFQ
GET	/rfq	List all RFQs
GET	/rfq/{id}	Get RFQ details and negotiation state
POST	/quote	Transporter submits a bid
GET	/transporters	List all registered LSPs
WS	/ws/{rfq_id}	WebSocket for live negotiation updates

Full interactive API docs available at /docs when running locally.

---

Environment Variables

Variable	Required	Description
GEMINI_API_KEY	Yes	Google Gemini API key from AI Studio

---

Frontend

The frontend is a separate project deployed on Vercel.

Live: v0-negotiation-agent-ui.vercel.app

It connects to this backend via REST and WebSocket. To point it at a local backend, update the API base URL in the frontend environment config.

---

Notes

• Data is currently persisted as JSON flat files in the /data directory. There is no database dependency — no setup or migrations required.
• The simulators/ module lets you run a full negotiation loop locally without needing real transporter logins, useful for development and testing.

---

Built at

LogisticsNow LoRRI Agentic Hackathon 2026