How Chokho AI Solves the Waste Management & Routing Problem
← Back to Blog

How Chokho AI Solves the Waste Management & Routing Problem

2026-06-09

Imagine taking an evening walk through a public park, only to find the pathways littered with scattered trash.

The immediate thought is usually frustration because it's unhygienic and ruins the environment. But let's say you're a responsible citizen who actually wants to do something about it. You decide to report it to the local municipality.

That's when you hit a wall of practical problems:

  • Who do you call?
  • Will they answer your call?
  • How will you explain the context and location?
  • Will they actually come and clean it?

Even if you manage to report it, will you again go through this just for a pile of trash you observed?

These were the exact reasons why I built Chokho AI. In pahadi, Chokho means pure/clean and it exactly aligns with our goal - to achieve a cleaner neighbourhood.

Chokho AI is a smart waste management and routing platform built for citizens to report trash and optimize the cleaning routes for municipal corporations.

High Level Architecture

The backend is built on a polyglot microservice architecture, where Spring Boot is used for buisness logic and FastAPI is used for ML operations, AI Inference and route optimization. Next.js is used for the frontend. For the database, I used PostgreSQL with PostGIS extension for spatial queries.

graph TD; subgraph Client Layer CitizenUI["Citizen Dashboard<br/>(React / Next.js)"] WorkerUI["Worker Dashboard<br/>(React / Next.js)"] AdminUI["Admin Dashboard<br/>(React / Next.js)"] end subgraph API Gateway / Core Layer SpringBoot["Core Backend<br/>(Spring Boot / Java)"] Auth["JWT Authentication<br/>& Role Management"] end subgraph Intelligence Layer FastAPI["AI Engine<br/>(FastAPI / Python)"] YOLO["YOLOv8n Model<br/>(Binary Detection)"] Gemini["Google Gemini AI<br/>(Validation & Metadata)"] TSP["Clustering & Routing<br/>(K-Means & TSP)"] end subgraph Data & Storage Layer Postgres[(PostgreSQL + PostGIS)] Cloudinary[(Cloudinary Image Storage)] end CitizenUI -->|RESTful APIs| SpringBoot WorkerUI -->|RESTful APIs| SpringBoot AdminUI -->|RESTful APIs| SpringBoot SpringBoot <--> Auth SpringBoot <-->|Geo-Queries & CRUD| Postgres SpringBoot -->|Image Uploads| Cloudinary SpringBoot <-->|Inference & Optimization Requests| FastAPI FastAPI --> YOLO FastAPI --> Gemini FastAPI --> TSP

Detailed Explanation

Phase 1 (Citizen Reporting & AI Validation)

When a user wants to report a trash, the system does the following operations:

  1. Firstly before letting user upload any image, the system makes sure that a user is not uploading an old image or any malicious file on our system by only allowing the user to directly open the camera from our system and upload the clicked image.

  2. When user hits the submit button, our system extracts the GPS coordinates from his browser and the image is uploaded. Then it goes through a finetuned YOLOv8n model to detect the location of trash.

  3. If trash is detected, then the image goes to the AI layer and it does the following things :

    • checks if the image is clicked in indoors
    • checks if the image is fake
    • measure the severity score (1.0 - 10.0)
    • checks the type of trash (PLASTIC, HAZARDOUS, DEBRIS, etc)
    • estimates the volume of trash (SMALL, MEDIUM, LARGE)
    • analyses the context of the image
    • analyses the location of complaint

  4. Once all conditions are met, the complaint is registered on the system, user gets notified and the complaint is publicly visible on the heatmap.

sequenceDiagram participant Citizen participant CoreAPI as Spring Boot participant AI as FastAPI Engine participant DB as PostgreSQL Citizen->>CoreAPI: Submit Photo + GPS Data CoreAPI->>AI: Request Initial Validation rect rgba(255, 255, 255, 0.1) Note over AI: YOLOv8 Validation AI-->>AI: Detect Trash Presence (best.pt) end alt No Trash Detected AI-->>CoreAPI: Validation Failed CoreAPI-->>Citizen: Error: Invalid Image else Trash Detected rect rgba(255, 255, 255, 0.1) Note over AI: Gemini AI Deep Analysis AI-->>AI: Extract Type, Volume, Severity AI-->>AI: Indoor/Outdoor Classification end AI-->>CoreAPI: Analysis Metadata alt Is Fake / Indoor CoreAPI-->>Citizen: Error: Fake / Invalid else Valid CoreAPI->>DB: Persist Complaint CoreAPI-->>Citizen: Success (Shows on Heatmap) end end

Phase 2 (Optimized Dispatch & Verified Cleanup)

Once a lot of complaints get registered on the system, they also get reflected on the heatmap. After taking a look at the heatmap, the admin can manually trigger the system to optimize the routes for the municipality vehicles.

  1. To optimize the routes, the system intially divides all the current active complaints into n clusters by using K-Means Clustering Algorithm. The value of n can be manually set by the admin or by default it is the number of available vehicles.

  2. Then for each cluster, the system implement TSP (Travelling Salesman Problem) on the data points and return a the optimized order of the points for that particular cluster.

  3. After calculating the optimized routes for all clusters, each cluster is mapped to a particular vehicle and that vehicle is assigned to a worker.

Everyday when a worker comes to his job, he can see his today's active route on his personalized dashboard.

  1. When a worker reaches a complaint location, he must send an after cleanup image of the same location to the system to verify the cleanup of that complaint.

  2. After a worker submits a cleanup image, that image goes through the AI layer and it verifies the cleanup image with the original complaint image.

  3. If the cleanup image is valid, the complaint's status is changed to CLEANED, heatmap gets updated and the user is notified of the cleanup with both complaint and cleanup image.

sequenceDiagram participant Worker participant CoreAPI as Spring Boot participant AI as FastAPI Engine participant DB as PostgreSQL participant Citizen Worker->>Worker: Arrive at location Worker->>CoreAPI: Submit Verification Photo + GPS CoreAPI->>DB: Query original complaint GPS CoreAPI->>CoreAPI: Calculate GPS Delta (< 20 meters) alt GPS mismatch CoreAPI-->>Worker: Error: You are not at the correct location else GPS match CoreAPI->>AI: Send Before & After Images AI-->>AI: Gemini visual comparison alt Images do not show cleanup AI-->>CoreAPI: Cleanup not verified CoreAPI-->>Worker: Error: Area still dirty else Verified AI-->>CoreAPI: Success CoreAPI->>DB: Update Complaint Status (Resolved) CoreAPI-->>Worker: Route Stop Completed CoreAPI-->>Citizen: Notification: Your report was resolved! end end

The Big Picture

Chokho AI Complete Architecture Overview

This is the complete architecture of how Chokho AI works! By combining the reliability of Spring Boot, the speed of FastAPI, and the intelligence of modern AI models, we can replace a broken, manual reporting system with an automated, optimized workflow.

Want to see it in action? Check out the demo video of Chokho AI here.

What's Next?

While the core architecture is solid, there is always room for improvement. My next goals for Chokho AI include:

  • Building a dedicated mobile app for citizens and workers for better user experience.
  • Expanding the AI model to recognize different classifications of waste (recyclable vs. non-recyclable) to optimize the type of truck dispatched.

Let's Connect!

I'd love to hear your feedback on the architecture! Feel free to connect with me on LinkedIn and let me know your thoughts.

If you are an investor, municipality representative, or developer interested in scaling this solution, I'd love to talk. Please reach out via email or send me a DM on LinkedIn.

Comments

0 Comments