How Raindrop became the Sentry of AI: Scaling to petabytes with Tinybird

About the company

Raindrop is building Sentry for AI products - a monitoring platform that gives companies visibility into how their AI applications are performing, what topics users are exploring, and where issues arise. AI engineers use Raindrop to discover AI app anomalies, set up tracking, and curate datasets from real user interactions.

Problem

Raindrop initially built their data infrastructure on Postgres, but when onboarding their first customer with millions of events a day, Postgres immediately started to fail. They faced poor performance with time series data, slow aggregation and filtering operations, difficulty handling text search at scale, and complex workflows around analyzing and processing events.

Why Tinybird

After switching from Postgres to Tinybird, Raindrop saw immediate performance gains - 100x to 1000x faster - even before any optimization. Tinybird's Playgrounds helped transfer Postgres knowledge to ClickHouse®, and the clean abstraction of pipes created separation between data and application layers.

Results

• 1 week to production. Switched from Postgres to Tinybird in about a week for their first POC customer.
• 100-1000x faster queries. Orders of magnitude faster even before optimization.
• 2-3 engineers saved. Avoided hiring dedicated ClickHouse engineers.
• Clean architecture. Pipes abstraction creates separation between data and application layers.

AI applications have been flying blind

As artificial intelligence becomes integrated into more products, companies face a new challenge: understanding how users interact with AI features and identifying when things go wrong.

Traditional software monitoring tools like Sentry have transformed how companies detect and respond to errors in their web applications. But the same paradigm doesn't exist yet for AI applications, leaving companies with limited visibility into their AI features' performance.

Raindrop is building Sentry for AI products—a monitoring platform that gives companies visibility into how their AI applications are performing, what topics users are exploring, and where issues arise.

First Postgres, then data exploded

As with many startups, Raindrop initially built their data infrastructure on Postgres. This worked well in the beginning, but quickly hit limitations as they started handling customer data.

The team faced multiple challenges with their Postgres implementation:

• Poor performance with time series data
• Slow aggregation and filtering operations
• Difficulty handling text search at scale
• Complex workflows around analyzing and processing events

They needed a solution that could not only support their user-facing analytics but also their internal metrics.

Raindrop chose Tinybird for speed and simplicity

Raindrop knew they needed a database designed for analytics workloads. After evaluating options like ClickHouse Cloud and various providers, they selected Tinybird for two critical reasons: implementation speed and reduced complexity.

Massive performance gains

After switching from Postgres to Tinybird, Raindrop saw immediate performance gains even without any optimization.

This immediate performance gain made the decision clear for Raindrop, but it wasn't just raw query speed that won them over. As the team began working with Tinybird, they discovered several key architectural advantages that aligned perfectly with their engineering values and security requirements.

Intuitive developer experience

For a small team transitioning from Postgres to a column-oriented analytics database, the learning curve could have been steep. Tinybird's developer-friendly interface helped bridge this knowledge gap.

The interactive SQL environment allowed Raindrop's engineers to experiment with queries, see immediate results, and understand the performance implications of different approaches. This accelerated their learning and enabled them to quickly build expertise in optimizing time-series analytics queries.

Clean architecture

Raindrop valued Tinybird's clean separation of concerns, which allowed them to maintain a well-structured application architecture while integrating powerful analytics capabilities.

This architectural elegance meant Raindrop could maintain a clear separation between their data processing logic and application code. Engineers could modify analytics queries without touching application code, and vice versa, reducing coupling and simplifying maintenance as the product evolved.

From Postgres to production in one week

Raindrop was able to move remarkably quickly with Tinybird, transitioning their first proof-of-concept customer from Postgres to Tinybird in about a week.

Tinybird's branching feature proved crucial for Raindrop's rapid development cycle, allowing them to experiment with different approaches without disrupting production.

This rapid implementation enabled Raindrop to build several key features powered by Tinybird:

Powerful filtering and aggregations

Tinybird makes it possible to narrow down a search to a specific property, for example, to see that an issue is primarily impacting a specific AI model or tool call.

Internal analytics & monitoring

Much of Raindrop's internal monitoring is built directly within Tinybird as well. By bringing Raindrop's internal analytics dashboards in-house, Raindrop can build powerful dashboards and alerting systems that can notify Raindrop and their customers, as well. For example, if a customer has a burst of 100k events, Raindrop's issue detection can experience a slight lag. Now with Tinybird, they can show that to their customers.

Time series analytics

Throughout the product – whether in Slack or the Dashboard – Raindrop shows customers charts to understand how usage of their AI product is changing over time. These graphs are all built on top of Tinybird. Powerful features like WITH FILL/STEP abstract the hardest parts of building reliable and performant charts.

Saving the equivalent of multiple engineering hires

When asked about the return on investment from implementing Tinybird, Hylak's answer was clear and quantifiable:

For a small startup team, this represents a significant cost saving and allows them to focus on building their core product rather than managing infrastructure.

Raindrop has also found a balanced approach between traditional databases and analytics platforms:

This practical approach has allowed Raindrop to build a robust monitoring platform that can scale with their customers while maintaining a lean engineering team.

Advice for teams evaluating similar challenges

For teams facing similar challenges with time-series data in Postgres, Hylak offers clear advice:

With Tinybird, Raindrop has built a platform that can process hundreds of millions of events daily, providing AI companies with the monitoring and visibility they need to build better AI products. The infrastructure is now positioned to scale with Raindrop's rapid customer growth while maintaining the performance and reliability that their customers depend on.