Conversations with Banyan Infrastructure's Engineering: First Few Months as an AI Transformational Architect

We spoke with Joshua Doan, AI Transformation Architect, who is tasked with leveraging AI within our product and internal operations. Here are some highlights that capture the impact and momentum of his work so far, in his own words.
🚀 AI Velocity Tracking
One of the foundational projects I led was the development of an AI-powered application that enables us to visualize and track engineering velocity. With the rise of AI-assisted development environments (such as Copilot and other tools), we needed a way to determine if these tools were truly making our team more efficient.
We built a platform that pulls data from YouTrack and GitHub, examining how long issues remain in progress, how long PRs (Pull requests) take to review and merge, and other key metrics. Then, using an LLM (Large Language Model), the system generates helpful charts and summaries that clearly illustrate our trends in development speed over time.
This tool enables our leadership team to assess whether the AI tools we've implemented are making an impact. It also provides a data-driven approach to refine and evolve our ongoing AI strategy. With this level of visibility into what works and what doesn't, we can confidently invest further in tools and practices that improve velocity without sacrificing quality.
đź§ Engineering AI IDE Tools
We also focused on helping our engineers maximize the benefits of AI directly within their development environments. We partnered with both frontend and backend teams to evaluate and implement in-IDE AI tools tailored to their workflows. Additionally, we rolled out Model Context Protocol (MCP) tools that allow seamless synchronization between our issue tracking (YouTrack), GitHub, and Figma systems. We also added dedicated context files to each repository, providing explicit, project-specific knowledge that the AI can use to deliver more accurate suggestions without requiring the engineer to type the same context repeatedly.
🤖 AI Review Bot with Contextual Awareness
Another breakthrough was the development of our AI Review Bot. Unlike most AI-powered code reviews, this bot understands both the product context (via YouTrack) and the code history (via GitHub). The team's feedback has been overwhelmingly positive. A favorite moment? One PR had the AI review bot, our Engineering Director, and a senior engineer all recommend the same fix for a bug simultaneously. That level of alignment shows just how well the bot is learning and integrating into our review workflows.
📝 Automated AI-Powered Release Notes
Creating detailed, product-relevant release notes used to be a tedious, manual process. Now, our AI system reviews each sprint and releases data to generate rich, polished summaries. These include:
- Epic-level highlights
- Key stories completed, prioritized by product impact
- Detailed, readable release summaries
All of this is automatically exported to our team wiki and even announced in Slack. It's a simple, repeatable, and automated flow that saves time and increases visibility for the entire company.
🛠️ AI-Powered Refactor Bot
One inspiring initiative was building a Refactor Bot capable of handling large-scale refactoring tasks across many files. Inspired by a great piece from Airbnb's engineering blog, we designed a system where the AI refactors each file individually based on a given prompt. Then, a separate verification prompt checks the output for quality and correctness, catching issues and retrying as needed. One notable use case: we used it to add automation test attributes across our frontend codebase at scale. This tool significantly reduces the manual effort required for broad, repetitive code updates.
These are just a few of the initiatives we've launched. Still, they represent a core theme: empowering our teams with AI in ways that are practical, measurable, and deeply integrated into existing workflows.
And the results so far speak for themselves:
- We've seen a 65% decrease in average development time since adopting AI-assisted workflows.
- Pull requests are getting merged nearly twice as fast.
- Our automated release notes are saving each product manager 2–3 hours every week, time they now spend on strategic planning and customer engagement.
- The Refactor Bot handled 100% of updates across more than 4,000 files in minutes, enabling sweeping, consistent code changes that would have taken days manually.
By grounding our strategy in data and real outcomes, we're building a foundation for long-term impact, and the momentum is only picking up. Empowering our teams with AI in ways that are practical, measurable, and deeply integrated into existing workflows.
What is next for AI at Banyan Infrastructure
As we look ahead, we have several exciting projects on the horizon:
- AI-Driven Product Insights: We're exploring ways to apply AI to our application analytics to help the product team uncover opportunities for improving the user experience and making our product even more delightful.
- Accelerated Onboarding Through AI-Powered Data Import: We're working on dramatically reducing onboarding time by using AI to import and organize large volumes of unstructured data into our system, removing a major bottleneck in customer setup.
- AI-Powered End-to-End Testing: Manually maintaining E2E testing is time-consuming and fragile. We're developing a solution that leverages AI to intelligently generate and maintain E2E tests—bringing speed and reliability to our quality assurance workflows.
- Friction Reduction Across the Organization: One of our guiding principles is to continually explore how AI can reduce friction, not just in engineering but across every department. From PR to customer success, we believe AI can help everyone move faster and smarter.
By thoughtfully integrating AI where it provides real value, we're accelerating the pace at which ideas move from concept to customer. As our tooling and processes continue to evolve, I'm excited to see just how far we can push the boundaries of speed, quality, and innovation.
‍