This project explores the impact of Artificial Intelligence (AI) on the performance and behavior of various Python web API frameworks. We evaluate how frameworks like Flask, FastAPI, and Django perform under load and how they integrate with lightweight, open-source AI models.
The goal is to understand:
- How API frameworks handle AI-related workloads
- What performance trade-offs occur when AI features are embedded
- Which framework is most suitable for lightweight AI-enhanced applications
What is the effect of AI on API frameworks in Python?
We focus on performance (latency, memory, request throughput) and developer experience when integrating AI tasks such as text summarization or sentiment analysis.
We compare three major Python web frameworks:
- Flask – Minimalistic and synchronous
- FastAPI – Async-based and modern
- Django – Full-featured and monolithic
Each framework serves a simple blogpost API with optional AI-powered features.
To simulate real-world use of AI in APIs, we implemented the following endpoints:
- Text summarization of blog posts using HuggingFace's
bart-large-cnn - Sentiment analysis using
bert-base-multilingual-uncased-sentiment - Title generation with GPT-like models
- Tag prediction using custom-trained classifiers
These tasks are integrated using open-source transformer models via HuggingFace Transformers.
We benchmarked each framework with and without AI features using the following metrics:
- Requests per second
- Memory usage
- Average latency
- Total runtime under load
- FastAPI consistently delivered the highest request throughput, but showed increased CPU usage and latency when handling concurrent AI workloads.
- Flask had the lowest memory usage and moderate performance, but less suited for async AI tasks.
- Django showed the highest memory footprint and slowest performance, but benefits from built-in tools and structure for complex apps.
Lukas Van der Spiegel
🌐 lukasvanderspiegel.be
🔗 linkedin.com/in/sidge
This project is licensed under the MIT License.