Arsalan Bashir

Arsalan Bashir founded LiveDocs, a data analysis notebook, and later built AI debugging tools at Sentry. At LiveDocs, he and his team rebuilt a no-code analytics product after customers kept bringing them questions that dashboards couldn’t answer on their own. At Sentry, he wrote changes that help an agent investigate what happened when a production metric crossed an alert threshold.

Why LiveDocs moved beyond dashboards

Arsalan trained as a mechanical engineer and worked in automotive R&D before moving into software. He founded and led LiveDocs, part of Y Combinator’s Winter 2022 batch. The team spent nearly two years building a no-code business intelligence tool before Arsalan concluded they had started too far downstream. Customers still had to locate their data, work around broken exports, and agree on what a metric meant. Requests for new metrics and integrations kept turning into one-off analysis for individual customers.

The team took a different approach with LiveDocs 3.0. They started with a prototype that could run terminal commands, then added web search so the data agent could find datasets and documentation. It could also use PDFs and screenshots alongside tables. A customer could start with the material they already had, before it was ready to chart.

How Sentry investigates a metric alert

As a staff software engineer working on AI at Sentry in 2026, Arsalan changed how metric alert investigations gather evidence. He gave query steps the resolved metric issue, monitor, project, and analysis window up front. The investigation could then examine the monitor’s context and the shape of the spike, look for contributing issues and shifts in tags, and identify likely triggers. Once the automatic steps finished, the workflow saved a short summary and an actionable description.

Arsalan also made investigation steps respect their dependencies. A step that uses earlier results now waits until all of those results are complete and still current. He made the latest usable query result available for refinements, too: when a developer asks to turn a line chart into a bar chart, the agent can redraw it with the data it already fetched. His scheduled AI Engineer Code 2026 session, Let the Agent Cook: From APIs to Answers, focuses on how agents discover tools, follow leads across services, and decide when they have enough evidence.