From Notebooks to Production: The Hard Truth About Shipping Your First AI App

2026-03-13 · Source: Towards AI - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cloud Computing & IT Infrastructure · Depth: Intermediate, long

Summary

An AI developer recounts a ten-month journey from building an AI model in a Jupyter notebook to shipping a full end-to-end application with a UI and real infrastructure. The initial approach, relying solely on Python frameworks like Streamlit/Gradio for the frontend and Flask for the backend, resulted in an unusable UI and deployment challenges. The author navigated multiple deployment platforms, including Railway, Google Cloud Run, Digital Ocean, and ultimately Render, to find stability. Resource management became critical, with configurations like 4GB RAM, 10GB storage, and strict artifact retention policies. The architecture evolved from two separate services to a consolidated single instance, requiring a shift from Python-centric UI tools to learning React in 3-4 weeks to achieve a production-ready frontend. Finally, implementing Google OAuth with 2FA established trust and security for the application.

Key takeaway

For ML engineers or backend developers building AI applications, recognize that your MVP stack will likely not be your production stack. Be prepared for architectural rewrites and invest in frontend development (e.g., React) and robust authentication early. Your deployment platform choice significantly impacts iteration speed and cost, so choose wisely to avoid operational overhead and focus on product features.

Key insights

Shipping an AI application demands diverse skills beyond model development, especially in UI, deployment, and architecture.

Principles

• UI is the product, not an afterthought.
• Deployment is a product decision.
• Constraints clarify technical thinking.

Method

The author iteratively refined an AI application by starting with Python-native tools, then migrating to React for the frontend, consolidating backend services, and implementing robust authentication and resource management.

In practice

• Prioritize UI/UX from early stages.
• Evaluate deployment platforms for iteration speed.
• Implement strong authentication early.

Topics

• AI Application Deployment
• Frontend Development
• Cloud Resource Optimization
• Software Architecture
• Authentication Systems

Best for: AI Architect, AI Engineer, Entrepreneur, Machine Learning Engineer, Software Engineer, MLOps Engineer

Related on AIssential

• How AI Companies Are Deploying Products at Enterprise (And the Role Making It Happen) — Enterprise AI success hinges on Forward Deployed Engineers who bridge the gap between product demos and complex customer deployment realities.
• How to Deploy Your First App on FastAPI Cloud — FastAPI Cloud streamlines web application deployment, offering a managed platform for FastAPI projects.
• Thinking in Tradeoffs: What to Get Right Before You Ship Your First AI Feature — Shipping an AI feature, particularly for resource-constrained startups, necessitates a deliberate approach to tradeoffs to prevent long-term production issues.
• Issue #121 - How to Deploy your AI Agent? — Deploying AI agents requires a comprehensive understanding of distributed systems, memory management, and operational challenges.
• How do companies decide between building AI models in-house or using APIs? — Companies face a critical build vs. buy decision for AI models.
• State of the Art of Platform Engineering • Abby Bangser & Charles Humble • GOTO 2026 — Platform engineering must provide self-service, API-first, deterministic capabilities to enable scalable, compliant AI agent consumption.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Towards AI - Medium.