11 CRAZY Backend Project Ideas That Pushes Your Limits

Overview

The video presents 11 backend project ideas designed to go beyond basic CRUD apps. They help developers demonstrate skills in real-time systems, data processing, AI integration, scalability, and infrastructure. The narrator alternates between approachable and ambitious ideas, and suggests adjusting each project’s scope to match your experience.

Project Ideas

1. Search Engine

Collect content by crawling websites or using public APIs. To keep the project focused, you could limit it to a single site, such as Reddit.

Process the content by cleaning HTML, normalizing text, removing stop words, and tokenizing it. Rank results using an approach such as TF-IDF, BM25, or vectorization.

Possible extensions include pagination, rate limiting, sharding, caching, background reindexing, prefix-tree autocomplete, and search analytics such as popular queries and click-through rates.

2. Skribbl-Style Multiplayer Drawing Game

Players draw assigned words while others guess; faster guesses earn more points. Building the game offers experience with WebSockets, game rooms, event-driven systems, chat, and scaling.

Variations could include having everyone draw the same prompt and vote on the best result, or building another simple multiplayer game.

3. Food-Delivery App with Short Videos

Let users discover food through short clips and swipe to order dishes they like. For an initial prototype, orders could notify restaurants to handle delivery themselves or redirect users to another ordering app.

The central product idea is to make food discovery more engaging by showing food being prepared and eaten, rather than relying only on marketplace photos.

4. AI Workflow Automation Platform

Create a low-code or no-code visual designer where users connect blocks and third-party integrations to build generative AI workflows. Example workflows might create and email meeting materials or use multiple agents to improve a résumé.

Users could save workflows as reusable templates, run them as background processes, expose them through an API, or use them in a browser. A knowledge base could support retrieval-augmented generation (RAG).

The narrator emphasizes that integrations and modular, scalable infrastructure are major challenges; the core pipeline logic may be less difficult.

5. “LLM Minecraft”

This is a smaller, game-like take on AI workflow automation. Users deploy multiple LLM instances, configure them, and optimize their responses in parallel.

Pipelines can be represented as directed acyclic graphs, with shared memory and a knowledge base for RAG. Start with API wrappers and dynamic pipeline logic, then add authentication, integrations, and other platform features as needed.

6. Banana Trading Company

Build a market game in which players invest starting currency in assets whose prices shift based on demand, algorithms, and chance.

Design rules for issuing player-created assets, ownership shares, supply limits, and price tracking. Create a pricing system that prevents wealthy players, or “whales,” from causing runaway money creation. Balance fairness and rewards to keep the game engaging; possible additions include a leaderboard and an in-game newspaper.

7. Event Photo Finder Using Facial Recognition

An organizer uploads event photos to a private space. A background pipeline identifies distinct faces and links them to the photos in which they appear.

A participant logs into that space, takes a photo with their front camera, and receives matching event photos. Key engineering concerns include background processing, latency, and keeping server costs low.

8. AI-Powered Search Engine

Extend a basic search engine with semantic search using vector embeddings instead of relying only on BM25. The narrator names vector database options including PGVector, Pinecone, ChromaDB, and Qdrant.

Retrieve relevant material and provide it, along with the user’s query, to an LLM to generate a context-grounded answer, forming a RAG system. More advanced systems can generate multiple searches from one query and recursively inspect linked pages. Result quality depends heavily on the search and retrieval approach.

9. Route Planner for Lower Pollution

Notify users when pollution along a saved route is comparatively low. Store user and route information, collect AQI, weather, traffic, and related data, and assess pollution along routes.

A background workflow can reevaluate routes roughly every 15 minutes and send alerts by phone, email, or WhatsApp. More advanced features could recommend longer, lower-impact routes and support runners and cyclists. The narrator particularly recommends this as a hackathon project.

10. Pull-Request Checker

Build a GitHub app that reviews pull requests for bugs, security vulnerabilities, inconsistent conventions, and potential improvements.

The project provides experience with GitHub OAuth and API integration, as well as designing a system that can handle many requests and accounts. Instead of relying on a single general-purpose LLM call, coordinate specialized model tasks—for example, reviewing readability, maintainability, minor issues, and security.

11. Managed Deployment and Infrastructure Platform

Start with a basic service that deploys static sites to an S3 bucket rather than attempting to reproduce a full platform immediately.

Accept a repository or archive, build command, and source directory; authenticate users, create deployment records, and queue deployment tasks. Use a scheduler, queues, workers, and retry logic.

Run each build in a Docker container: clone the repository, install dependencies, build the project, copy the output to an S3 location identified by the deployment ID, and destroy the container. Add a proxy to route traffic to the correct deployed files. The narrator presents this as a simplified version of the architecture behind platforms such as Vercel.

Broader Advice and Sponsor Segment

These projects are intended to demonstrate capabilities beyond CRUD, but their scope can be adjusted. Some can be built as small prototypes, while others can grow into substantial systems.

The narrator recommends creating a detailed project plan before starting and suggests that viewers can generate one from the video transcript.

A sponsored segment promotes Scrimba, describing its project-based web-development learning, backend topics such as networking, Git, cybersecurity, and DevOps, and an interactive player where learners can edit and run the code shown.

Speakers and Sources Featured

• Manware — the video’s narrator and presenter.
• Scrimba — sponsor mentioned in a promotional segment; no separate sponsor speaker is identified in the subtitles.
• Lovecraft community on X — cited as inspiration for the “LLM Minecraft” concept.
• Other named organizations and products — Mozilla, Cloudflare, LangChain, GitHub, Vercel, and several vector databases are mentioned, but the subtitles do not identify their representatives as speakers.