Artie: Real Time Data Streaming For The AI Age
Jacqueline Cheong and Robin Tang, founders of real-time data streaming platform Artie, discuss their journey from identifying the critical need for fresh data at companies like OpenDoor to building a production-ready solution, acquiring their first major customer Substack via a cold email, and navigating the complex technical challenges of real-time data processing at scale.
Artie is a real-time data streaming platform designed to help companies move data across their systems as changes happen. For instance, it can stream production data from a Postgres database into a data warehouse like Snowflake in real-time. The company recently announced a $12 million Series A funding round.
The Problem: The Need for Real-Time Data
The idea for Artie originated from co-founder Robin Tang’s experiences at companies like OpenDoor and Zendesk. A recurring challenge was the inability to get fresh, up-to-date data from the data team. This bottleneck hindered crucial business functions:
- Faster Experimentation: At OpenDoor, faster data was needed to accelerate growth experiments.
- Operational Use Cases: Building internal tools with platforms like Retool was difficult without real-time data access.
Data teams often dismissed these requests as too difficult or resource-intensive unless they were a top company priority. Existing managed batch ETL tools were good for SaaS sources like HubSpot but were geared towards a different buyer persona (business users) and couldn’t handle the load from production databases, which are the domain of infrastructure and platform engineers.
An attempt to build an in-house solution to stream data from Postgres to Snowflake at a previous company failed to become production-ready even after a year of development. This highlighted a common, nonsensical pattern: companies were forced to spend years building data connectors, which is not their core competency, just to enable their primary business functions.
Go-to-Market: Acquiring the First Mission-Critical Customer
Artie’s product is critical infrastructure; if it breaks, it’s a major issue for the customer. This makes the sales cycle challenging, as it requires a high degree of trust from large companies.
- First Customer: The first major customer was Substack, who needed a reliable system to get data out of a massive Postgres database and into Snowflake with low latency, without impacting their production application’s performance.
- Acquisition: Substack was acquired through a cold email. The need was so acute that the head of data responded within an hour and scheduled a call for the same afternoon.
- Building Trust: Despite Artie being an unproven product, Substack gained confidence through a proof-of-concept where they pushed billions of rows and set tight constraints, which the system handled successfully. The alternative for them was a multi-year, multi-person internal project with a high risk of failure.
The journey was slow initially. After landing Substack quickly, it took another nine months to secure a second customer of a similar scale.
Lean Growth and Founder-Led Sales
Artie has been able to achieve significant milestones with a very small team, following YC’s advice to be cautious and deliberate about hiring.
- $1M ARR with 4 People: The company crossed the $1 million ARR milestone with just the two co-founders and two engineers.
- Founder-Led Growth: Co-founder Jacqueline Cheong handled all sales herself initially. This kept the feedback loop between the customer and the engineering team extremely tight, allowing for rapid product iteration. Feature requests and feedback were often implemented surprisingly quickly because the founders were on the front lines of both sales and development.
Technical Deep Dive: The ‘Accumulated Battle Scars’ of Data Processing
Building a reliable, scalable real-time data platform involves solving numerous difficult technical challenges. The founders describe data processing as “a series of accumulated battle scars” where experience with edge cases prepares you for the next one. A solution that works locally can fail spectacularly at production scale and complexity.
Key technical hurdles included:
- Handling Data Quirks: Dealing with unexpected and illogical data formats from sources, such as MongoDB allowing a month value greater than 12.
- Efficient Backfills: The initial backfill method, which streamed live changes while performing a historical sync, was too slow for massive tables (e.g., a 10-billion-row table was estimated to take 2.5 months). This required developing highly efficient, source-specific methods, such as querying underlying system columns in Postgres to extract data at I/O speed rather than through the SQL engine.
- Undocumented Features: To build a performant SQL Server connector using Change Data Capture (CDC), the team had to implement a non-standard, undocumented “backdoor method” because the officially recommended approach (backed by database triggers) was not performant enough for enterprise use.
- Third-Party Bugs: The team spent a year debugging an out-of-order data issue that was ultimately traced back to a bug in the Kafka SDK they were using. The library did not correctly implement the Kafka improvement plan for consumer rebalancing, causing data to be read out of order. This highlights a key challenge: “Their bug is your bug,” and you are responsible for the entire data integrity chain.
- Customer-Specific Architectures: The product must handle complex scenarios like sharded databases and multi-tenant architectures (e.g., ingesting and merging data from 20,000-30,000 individual databases into unified tables) without breaking ordering guarantees or schema evolution.
The Future: Scaling to Trillions of Rows and Beyond
Artie has processed over 700 billion rows of data in the last 12 months, a 12x increase from the prior year. With the rise of AI and agentic workloads demanding real-time data, the focus is on enhancing reliability and scalability to handle trillions of rows.
Product Roadmap:
- Beyond CDC: The company has launched an Events API to stream event data into warehouses like Snowflake and Databricks with latencies as low as 100-200 milliseconds. This leverages new native streaming capabilities being released by the major data warehouse platforms (e.g., Snowpipe Streaming).
- New Destinations: The roadmap includes expanding to destinations like Elasticsearch to power real-time search indexing directly from database changes.
To support this growth, the company plans to triple its team in the coming year, hiring across engineering, sales, marketing, and operations.