Build Real Things

From idea to portfolio project.

12 AI/ML project ideas with suggested tools, skills to practice and estimated effort. Open a build brief to plan your own implementation. These are project ideas, not downloadable finished applications.

  1. 01Define the problem
  2. 02Build a small prototype
  3. 03Evaluate the results
  4. 04Document & demonstrate

2

Beginner

5

Intermediate

5

Advanced

4

Categories

12 projects
Intermediate

AI-Powered Resume Screener

LLM that ranks resumes against job descriptions

Build a tool that takes a job description and a batch of resumes and outputs a ranked shortlist with reasoning. Uses LLM-based structured extraction and scoring.

GPT-4oLangChainFastAPIStreamlit

Why this project

Portfolio-ready tool recruiters actually want

~12h estimated · Project idea
Open build brief

Goal: LLM that ranks resumes against job descriptions.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using GPT-4o and LangChain. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: Prompt engineering, Structured output, Python.

Discuss your implementation
Beginner

Podcast Summarizer + Chapter Generator

Transcribe audio and produce structured summaries

Use Whisper to transcribe audio, then an LLM to extract key points, chapters, and a one-paragraph TL;DR. Add a web UI for drag-and-drop uploads.

WhisperGPT-4o-miniGradioPython

Why this project

Real-world audio AI pipeline in 6 hours

~6h estimated · Project idea
Open build brief

Goal: Transcribe audio and produce structured summaries.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using Whisper and GPT-4o-mini. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: Audio processing, Prompt chaining, File handling.

Discuss your implementation
Intermediate

Semantic Search Engine for Docs

RAG over your own document library

Build a semantic search system over PDFs or Notion pages. Embed documents into a vector store, then retrieve + rerank relevant chunks and generate grounded answers.

OpenAI EmbeddingsChromaDBLangChainStreamlit

Why this project

Production-ready RAG system - directly hirable skill

~16h estimated · Project idea
Open build brief

Goal: RAG over your own document library.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using OpenAI Embeddings and ChromaDB. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: Embeddings, Vector search, RAG, Chunking.

Discuss your implementation
Advanced

Multi-Language Sentiment Dashboard

Classify sentiment from tweets in 10+ languages

Stream Twitter/X data, run multilingual sentiment classification using a fine-tuned XLM-R model, and visualize trends in a live dashboard.

XLM-RoBERTaHuggingFacePlotlyFastAPI

Why this project

Shows you can work with real-time data at scale

~24h estimated · Project idea
Open build brief

Goal: Classify sentiment from tweets in 10+ languages.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using XLM-RoBERTa and HuggingFace. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: HuggingFace Transformers, Multilingual NLP, Streaming.

Discuss your implementation
Intermediate

Real-Time Object Detection App

Webcam inference with YOLOv8 in the browser

Run YOLOv8 on live webcam feed in the browser using ONNX Runtime Web. No server required - pure client-side ML inference.

YOLOv8ONNX Runtime WebNext.jsTailwindCSS

Why this project

Impressive live demo - perfect for interviews

~10h estimated · Project idea
Open build brief

Goal: Webcam inference with YOLOv8 in the browser.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using YOLOv8 and ONNX Runtime Web. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: Model export (ONNX), Web ML, JavaScript.

Discuss your implementation
Advanced

Medical Image Classifier

Fine-tune ResNet on chest X-rays

Fine-tune a ResNet50 on the NIH Chest X-Ray dataset to classify 14 pathologies. Implement Grad-CAM for explainability.

PyTorchResNet50Grad-CAMWeights & Biases

Why this project

High-impact domain AI - stands out in healthcare roles

~30h estimated · Project idea
Open build brief

Goal: Fine-tune ResNet on chest X-rays.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using PyTorch and ResNet50. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: Transfer learning, PyTorch, Medical imaging, XAI.

Discuss your implementation
Intermediate

AI Photo Restoration Tool

Restore old and damaged photos with diffusion

Use a combination of inpainting diffusion models and super-resolution to restore scratched or faded photos. Build a simple web UI around it.

Stable DiffusionGFPGANGradioPython

Why this project

Viral-worthy output - great for showing on social

~14h estimated · Project idea
Open build brief

Goal: Restore old and damaged photos with diffusion.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using Stable Diffusion and GFPGAN. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: Diffusion models, Inpainting, Super-resolution.

Discuss your implementation
Beginner

AI Writing Coach

Critique and rewrite text in different styles

Build a writing assistant that analyzes text for clarity, tone, and grammar - then rewrites it in the user's target style (formal, casual, persuasive). Stream outputs.

Claude 3.5 SonnetVercel AI SDKNext.jsshadcn/ui

Why this project

Deployable in a day - real users, real usage

~8h estimated · Project idea
Open build brief

Goal: Critique and rewrite text in different styles.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using Claude 3.5 Sonnet and Vercel AI SDK. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: Streaming, System prompts, UI/UX for AI.

Discuss your implementation
Advanced

AI Agent with Tool Use

Build an agent that can search, code, and browse

Create a LangGraph or CrewAI agent with tools: web search, code execution, and file read/write. Give it a task like 'research X and write a report' - watch it plan and execute.

GPT-4oLangGraphTavily SearchPython

Why this project

Agents are the hottest skill in 2026 AI hiring

~20h estimated · Project idea
Open build brief

Goal: Build an agent that can search, code, and browse.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using GPT-4o and LangGraph. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: Agents, Tool calling, LangGraph, Async Python.

Discuss your implementation
Intermediate

Personal Knowledge Base Chatbot

Chat with your Notion, Obsidian, or PDF notes

Connect your personal notes, sync them into a vector database, and build a conversational interface that retrieves relevant context and answers questions grounded in your knowledge.

OpenAISupabase pgvectorNext.jsLangChain

Why this project

Genuinely useful product you will use daily

~18h estimated · Project idea
Open build brief

Goal: Chat with your Notion, Obsidian, or PDF notes.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using OpenAI and Supabase pgvector. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: RAG, Note parsing, Chat UI, Embeddings.

Discuss your implementation
Advanced

ML Model Monitoring Dashboard

Track model drift, latency, and error rates

Deploy a model and wrap it with a monitoring layer that tracks prediction distribution drift, latency percentiles, and error rates. Trigger alerts when drift exceeds a threshold.

FastAPIEvidently AIGrafanaPostgreSQL

Why this project

MLOps is the #1 gap in most ML engineers' portfolios

~22h estimated · Project idea
Open build brief

Goal: Track model drift, latency, and error rates.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using FastAPI and Evidently AI. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: Statistical drift detection, Observability, FastAPI.

Discuss your implementation
Advanced

Fine-Tune Llama with LoRA

Fine-tune a 7B LLM on custom data in an hour

Use QLoRA to fine-tune Llama 3 8B on a custom instruction-following dataset. Push to HuggingFace Hub and deploy via Ollama for local inference.

Llama 3 8BQLoRAHuggingFaceOllama

Why this project

Fine-tuning skill is rare and extremely well-paid

~10h estimated · Project idea
Open build brief

Goal: Fine-tune a 7B LLM on custom data in an hour.

  1. Define the inputs, expected output and a small representative set of test cases.
  2. Build the smallest working version using Llama 3 8B and QLoRA. Keep credentials outside your code.
  3. Evaluate quality, failure cases and runtime cost. Document limitations before sharing your demo.

Practice: LoRA/QLoRA, HuggingFace PEFT, VRAM optimization.

Discuss your implementation

Submit Your Project

Built something with AI? Share it with the Thrive With AI community. Get feedback, visibility, and connect with other builders.