SignalPilot, a US-based AI startup founded by Bangladeshi diaspora engineers, has become the first team in the world to surpass the 50% mark on the hardest benchmark in data engineering AI—a threshold that teams from OpenAI, Anthropic, Google DeepMind, and JetBrains had been unable to cross in the eleven months since the benchmark was established.

The open-source AI agent achieved 51.56% accuracy on the benchmark on 21 April, crossing the 50% ceiling for the first time in the test’s eleven-month history. JetBrains, which had previously held the top position, responded within ten days—a turnaround that the company says is itself an indicator of how seriously the result was received.

The benchmark tests AI agents against broken enterprise dbt pipelines, a scenario that closely mirrors what data teams encounter in production. Its difficulty stems from a core problem that undermines AI agents deployed on real databases: they guess at schemas, misread relationships, and in some cases attempt to execute destructive operations on live infrastructure.

SignalPilot is built as a Jupyter-native agentic tool for data teams, connecting directly to a data warehouse, dbt lineage, query history, Slack threads, Notion, and Jira to give the AI the kind of institutional context that general-purpose tools cannot access. Rather than single-shot code generation, it runs long-running agent loops that plan, execute, and iterate with an analyst in the approval chain. It retains memory across sessions—tracking past hypotheses, validated assumptions, and known data quirks—and can be taught a team’s custom business logic, coding standards, and domain-specific analysis patterns.

The underlying system, called AutoFyn, runs Claude in sandboxed loops and iteratively tunes its own agent architecture until it converges on a stable state. AutoFyn also autonomously discovered 26 zero-day vulnerabilities in major open-source projects during development. The product is fully open-source and deployable locally in 60 seconds, with over 100 developers having used it within its first week.

The company is led by Tarik Adnan Moon, who trained as a mathematician at Harvard and worked as a quantitative analyst at Goldman Sachs before turning to AI agent infrastructure. Co-founder Fahim Aziz is a two-time Y Combinator founder. The team operates from the United States and is targeting a market of approximately 80,000 data teams globally that depend on dbt, the open-source data transformation tool around which SignalPilot is built.