How Delphi ships 100 times a day with its Python backend on Vercel

Delphi, a platform that converts expertise into interactive digital minds, initially relied on a Python backend hosted on AWS with ECS, Docker Desktop, and local databases. The setup worked for a small team but became inefficient as Delphi grew, with new engineers requiring a full day to set up their environment before their first deploy. The backend’s complexity made rapid iteration difficult, prompting the team to seek a simpler solution.

To address these challenges, Delphi migrated its Python backend to Vercel, leveraging Vercel Workflows and Vercel Queues to handle long-running tasks and background jobs without additional infrastructure setup. The move eliminated the need for staging deployments, allowing the team to ship changes directly to production up to 100 times a day behind feature flags and A/B tests. Vercel Agent’s anomaly detection further streamlined the process by identifying issues in real time.

The shift to Vercel also enabled non-engineering teams, such as growth and product teams, to build and deploy their own dashboards and experiments independently. Engineers now primarily interact with the codebase through agents, using Vercel’s SDK, MCP, and CLI to troubleshoot and deploy. Preview deployments provide live URLs for every push, allowing immediate feedback from any device or platform like Slack.

Delphi’s internal agents, including one built with Vercel’s agent framework eve, now handle tasks like diagnosing customer issues via Slack and running code in isolated environments. The platform also uses AI Gateway to dynamically route chat traffic to the best-performing AI models and manage failover, enabling same-day adjustments when new models are released. This infrastructure supports Delphi’s expansion beyond chat interfaces into features like multi-perspective search.