🛰️ ChainLLM

🔗 Natural Language to Geospatial Workflows via LLM Reasoning

A full-stack intelligent system that turns natural language queries into GIS workflows using chain-of-thought LLM reasoning, integrating MySQL, MongoDB, Spring Boot, FastAPI, and Ollama.

---

🚀 Overview

ChainLLM enables users to describe complex geospatial analysis tasks (e.g., “Map flood zones near Brahmaputra river within 10km radius of schools”) in plain English. It uses LLMs (like Mistral via Ollama) to generate step-by-step GIS workflows, passes them to a Python-powered GIS execution engine, and provides results—all tied to a unique taskId.

---

🧱 Architecture

graph TD
    A[Frontend Next.js] -->|/llm/ask| B[Spring Boot API]
    B -->|Generate taskId + trigger LLM| C[Ollama Mistral]
    C -->|LLM Response + taskId| D[FastAPI GIS Engine]
    D -->|Process GIS task| E[MongoDB]
    B -->|Track task metadata and Authentication| F[MySQL]
    E -->|Result files| G[Frontend polls/view]

Loading

---

⚙️ Tech Stack

Layer	Tech
Frontend	Next.js, TailwindCSS
API Gateway	Spring Boot (Java)
Auth/Task DB	MySQL
LLM	Ollama (locally hosted, e.g. Mistral)
GIS Engine	FastAPI (Python)
GIS Output DB	MongoDB

---

🌐 Flow Summary

1. User submits prompt on frontend (/llm/ask).

2. Spring Boot:

   • Generates and returns a unique taskId immediately.
   • Sends prompt to Ollama (LLM inference).

3. LLM Response is sent to the FastAPI GIS server with the same taskId.

4. Python GIS Engine:

   • Parses LLM output.
   • Executes GIS operations (GeoPandas, Rasterio, etc.).
   • Stores logs/output in MongoDB.

5. Frontend polls or fetches final results using taskId.

---

📦 Project Structure

chainllm/
├── frontend/             # Next.js frontend
├── backend/              # Java Spring Boot
│   ├── controller/
│   ├── service/
│   └── model/ (MySQL entities)
├── gis-engine/           # FastAPI Python service
│   ├── routes/
│   ├── services/
│   └── models/ (MongoDB schemas)

---

📊 Databases

System	Purpose
MySQL (Java)	User auth, taskId tracking
MongoDB (Python)	GIS output, logs, CoT steps

---

🛠️ Run Locally

🧩 Prerequisites Node.js (v18+ recommended) + npm → for running the Next.js frontend

Java 17+ and Maven → for the Spring Boot backend

Python 3.10+ and pip → for the FastAPI GIS engine

MySQL and MongoDB running locally → MySQL for Spring Boot, MongoDB for GIS logs

Ollama installed → run with ollama run mistral

Docker (optional) → for containerized setup (optional)

---

1. Start Spring Boot Server

cd backend
./mvnw spring-boot:run

---

2. Start FastAPI GIS Server

cd python GIS
uvicorn main:app --reload --port 8000

---

3. Start Frontend

cd frontend
npm install
npm run dev

---

📂 Sample Prompt

“Generate a flood risk map for Guwahati with buffers around water bodies and elevation analysis.”

➡️ Returns:

• Chain-of-Thought JSON
• GIS logs
• Downloadable raster/vector results

---

✅ Features

• 🔐 JWT-based login (Next.js + Spring Boot)
• 🧠 Real-time LLM response via Ollama
• 🛰️ Automated GIS analysis with reasoning logs
• 📈 Result tracking via taskId
• 💽 Dual database architecture

---

📚 Datasets

• OpenStreetMap (OSM)
• Bhoonidhi
• Sentinel-2, SRTM, etc.

---

🧠 LLM Setup (Ollama)

Install Ollama: https://ollama.com Run locally:

ollama run mistral

---

📥 API Endpoints

Endpoint	Method	Description
/llm/ask	POST	Submit prompt, get taskId
/api	POST	Called internally by Java app
/auth/login	POST	Auth endpoint
/api/task/:taskId/	GET	(Optional) fetch processed data

---

✨ Future Features

• ✅ Frontend dashboard with result previews
• 🔁 Retry failed GIS jobs
• 🗃️ Archive & replay prompt history
• 🧾 Visual Chain-of-Thought logs
• 📊 GIS metric comparison dashboard

---

👥 Team Ashwamedh

Name	Role	GitHub
Debashis Mitra	Team Lead, Java Backend, DevOps	@Debashismitra01
Debankur Dutta	Python Backend, GIS	@Debankur04
Susovan Sinhababu	Frontend Developer	@Susovan700
Anuska Kapuria	Documentation, PPTs & Presentation	@Anuska1312

---

License

© 2025 Team Ashwamedh. All rights reserved. This source code is proprietary and intended solely for evaluation purposes by the ISRO BAH committee. Any reproduction, distribution, or use without explicit permission is strictly prohibited.