EG
EngineerguideENGINEERGUIDE.COM
Start Learning→
APPLIED AI & DATA

AI & Machine Learning Projects

Move beyond Jupyter notebooks running pre-canned tutorials. High-value ML projects show that you can process raw data, handle vector embeddings, manage inference latency, and deploy production endpoints.

Coupled Learning Track
AI & Machine Learning Roadmap
View Roadmap Stages →

Why Build These Projects?

Engineering hiring managers look for engineers who understand model tradeoffs, vector retrieval, evaluation metrics, and how to interface ML models with real applications.

PROJECT 01·Advanced

Document Q&A Pipeline with Retrieval-Augmented Generation (RAG)

Build an end-to-end knowledge assistant that parses engineering documentation, creates semantic vector embeddings, and performs grounded retrieval with source attribution.

Prerequisites:Python intermediateVector math & cosine similarity conceptsREST APIs

Recommended Tech Stack

PythonFastAPILangChain or LlamaIndexChromaDB or pgvectorHugging Face / OpenAI Embeddings

Key Skills You Demonstrate

  • ✓Semantic text chunking strategies (recursive, semantic, windowing)
  • ✓Vector database indexing and top-K similarity search
  • ✓Prompt engineering for grounded answers without hallucination
  • ✓Evaluating retrieval quality with precision and faithfulness metrics

Core Features to Implement

1.Multi-format file ingestion (PDF, Markdown, HTML)
2.Hybrid search combining BM25 keyword search and vector cosine similarity
3.Exact paragraph citation links for every generated answer
4.Asynchronous background worker queue for document ingestion
💡 Architecture Tip

Do not use giant chunk sizes. Optimal retrieval typically uses 250-500 token chunks with 50-token overlaps.

Seniority Extension Ideas (To Stand Out in Interviews)
  • →Add a reranker model (like Cohere or BGE Reranker) to refine top-K candidates before prompting
Reference Docs:Stanford CS229 Lecture Notes↗Hugging Face Documentation↗
Back to AI & Machine Learning Roadmap →
PROJECT 02·Intermediate

Real-Time Computer Vision Classifier with Model Serving

Train a convolutional or vision-transformer model, optimize inference weights, and serve predictions via a low-latency REST and gRPC API.

Prerequisites:PyTorch basicsConvolutional network architecture intuitionDocker

Recommended Tech Stack

PyTorchTorchScript / ONNXFastAPIDockerOpenCV

Key Skills You Demonstrate

  • ✓Data augmentation and transfer learning techniques
  • ✓Exporting models to ONNX runtime for 3x CPU inference acceleration
  • ✓Batched tensor operations and image preprocessing pipelines
  • ✓Containerized model deployment

Core Features to Implement

1.Drag-and-drop web interface for testing image predictions
2.Real-time confidence scoring and top-3 class breakdown
3.Model inference time monitoring (latency p50 and p99 logging)
4.Dockerized production container under 1GB size
💡 Architecture Tip

Warm up the model upon container startup by passing a dummy zero tensor before opening the HTTP port.

Seniority Extension Ideas (To Stand Out in Interviews)
  • →Implement video stream processing with OpenCV tracking objects across frames
Reference Docs:Official PyTorch Tutorials↗
Back to AI & Machine Learning Roadmap →

Explore Other Project Domains

Web & Full-Stack Systems Projects →DevOps & Cloud Infrastructure Projects →Beginner & Student Portfolio Starters →