Open Source Startup Podcast

This episode dives into why code quality still matters in the age of AI, and why English - no matter how good models get - won’t replace programming.

Our guest is Co-Founder of Boundary, Vaibhav Gupta, and he shares the journey behind BAML, a new programming language to write and manage AI logic. After 12 pivots and 3.5 years, the team realized something simple but powerful:
AI tools were evolving fast, but the code was ugly.

Most AI generated code was unnecessarily long and messy. For builders who viewed code as artistic expression, that was painful. Once they tried BAML, everything changed. It was clean, elegant - completely the opposite of AI slop.

It wasn’t an overnight success. It took nine months to reach ten users — but the early ones stayed because of thoughtful design:

• Easy model swapping

• Full visibility into every prompt and test case

• A workflow so simple that non-technical users (even lawyers!) could test code

BAML was built with a philosophy that code is the source of truth, not the docs.

The conversation touches on how LLM observability and thoughtfully designed code make BAML unique. It’s inspired by the same thinking that made React sticky - beauty and composability.

Pretty code, the founder believes, isn’t vanity - it’s a functional advantage:

• Fewer bugs

• Easier to reason about

• Friendlier for AI-generated systems

In the episode, we sat down with ClickHouse Co-Founder Yury Izrailevsky to unpack how one of the fastest open-source databases in the world became the analytics engine of choice for 2,000 customers including Harvey, Canva, HP, and Supabase. From its Yandex origins to powering AI observability, Yury shares how ClickHouse balances open-source roots, cloud innovation, and a remote-first culture moving at breakneck speed.

ClickHouse's Series C valued the company at $6.35B earlier this year, and just yesterday they announced an extension to that round, just months after it was raised.

In this episode, we dig into:

• Origins & Founding Story

  • ClickHouse began as an internal project at Yandex to power a Google Analytics–style platform, focused on performance and scale.

  • Open-sourced in 2016 - rapid global adoption laid the foundation for ClickHouse the company.

  • Yury first discovered ClickHouse while at Google; impressed by its speed, he later co-founded the company in 2021 alongside Aaron Katz (ex-Elastic) and the original creator Alexey Milovidov.

• Why ClickHouse Stands Out

  • Column-oriented, open source OLAP database designed for massive-scale analytical processing.

  • Excels in performance, efficiency, and cost - ideal for large data volumes and real-time analytics (and now AI workloads).

  • Architectural choices:

    • Columnar storage = better compression and faster execution.

    • Separation of compute and storage enables elasticity, scalability, and resilience in the cloud.

• Open Source vs. Cloud

  • Open-source version offers freedom and flexibility.

  • Cloud product delivers much lower total cost of ownership and fully managed experience.

  • Architectural parity between the two ensuring no vendor lock-in for customers.

  • Customers can run the same queries on both; most stay with cloud due to simplicity and cost efficiency.

• Use Cases & Ecosystem

  • 4 main use cases:

    1. Real-time analytics

    2. Data Warehousing

    3. Observability

    4. AI / ML Workloads

• Company Building & Culture

  • Fully remote from day one.

  • Prioritized experienced, self-sufficient engineers over early-career hires.

  • Built and launched GA version in less than a year - insane pace of innovation.

• Innovation & Community

  • Monthly release cadence.

  • Hundreds of integrations and connectors.

  • Strong open-source and commercial community

• Advice for Founders

  • Focus on what matters most

  • Hire mature, independent thinkers.

  • Move fast but maintain quality; ClickHouse Cloud achieved production-grade quality in record time.

Chang She is Co-Founder & CEO of LanceDB, the multimodal lakehouse platform. Their open source data format lance has over 5K stars on GitHub and is a modern columnar data format for ML and LLMs implemented in Rust.

LanceDB has raised $41M from investors including Theory Ventures, CRV, and Essence VC.

In this episode, we dig into:

• Early focus: autonomous vehicles; solved real-time analysis limits with Lance format → 9,000% performance gain.

• Multi-modal AI taking off (vision, audio, text); Midjourney & Runway as pioneers; audio now a major category.

• How they built trust through open source.

• Integrated workflows (data prep + search + embedding) going beyond vector DBs; education needed to show full value.

• Cloud/serverless launch in 2023–24 enabled seamless local-to-production use.

• Future bets: audio infra, robotics, spatial reasoning; vector DBs risk irrelevance if they don’t evolve.

Ryan Djurovich is the Founder & CEO of Nadrama, the open source infrastructure automation platform that deploys containers instantly.

In this episode, we dig into:

• Kubernetes challenges that still exist today – setup and operations are notoriously hard and complex.

• What great developer experience means to him – focused on making deployments super simple by streamlining infrastructure and common tasks.

• Core value of Nadrama – developers just want to deploy apps; Nadrama abstracts away infrastructure pain.

• His view on what being truly open source means (including using the Apache 2.0 license)

• Ryan's user discovery process - talking directly with as many users as possible, mining his network / folks he's worked with in the past, community events & meetups.

• Navigating the earliest days of Nadrama

• Security philosophy – believes in baseline security for all accounts (not just enterprise), informed by a Cloudflare background.

Morgante Pell is the Founder of Grit, the developer tool that puts software maintenance on autopilot and was acquired by Honeycomb in April 2025.

In this episode, we dig into:

• The Grit product and how LLMs have made software maintenance much more efficient

• Launching GritQL - Grit's embedded query language for searching and transforming code

• Their early focus on the JavaScript community

• The motivation for Grit to open source

• How AI generated code has put pressure on software maintenance

• Where new problems have been created by AI-generated code

• The acquisition by Honeycomb - motivation, integration, and how the deal happened

Emma Burrows is Co-Founder & CTO of Portia AI, the platform to build AI agents in regulated environments. Their open source Python SDK provides a developer framework for predictable and stateful agentic workflows.

Portia AI has raised around $5M from investors including General Catalyst and First Minute Capital.

In this episode, we dig into:

• Why they built an end-to-end platform from agent planning to deployment

• The focus on accuracy as their true north star metric

• Their paid contribution program

• How their found their initial ICP in regulated industries

• Why 2026 will be the year of agents

• Her best fundraising advice (hint: never really stop fundraising)

Alan Braithwaite is Co-Founder & CTO of RunReveal, the security data platform with real-time monitoring, built-in detections, and AI-powered investigations. Today, they manage and analyze security logs for teams at Harvey, ClickHouse, Cloudflare, and Temporal.

RunReveal has multiple open source projects including event stream processing library kawa and query language pql.

RunReveal has raised from investors including Costanoa, Modern Technical Fund, and Runtime Ventures.

In this episode, we dig into:

• Why today's modern security teams are rethinking data management

• The benefits of building RunReveal on ClickHouse

• How they worked with early believers / customers like Temporal

• Their open source strategy and building trust with the community through open sourcing components like their event processing library

• Their MCP server and enabling security teams to use AI to automate investigations (including the launch of their new remote MCP server)

Vasek Mlejnsky is Co-Founder & CEO of E2B, the open-source runtime for executing AI-generated code in secure cloud sandboxes. Essentially, they give AI agents cloud computers.

Their open source repos, particularly e2b which has 9K GitHub stars, have been widely adopted to help securely run AI-generated code.

E2B has raised $12M from investors including Decibel and Sunflower.

In this episode, we dig into:

• Why agents need a sandbox

• Building a new category of infra tooling, much like LaunchDarkly

• Some of their viral content moments - including Greg Brockman sharing their videos

• Figuring out the right commercial offering

• Why they don't agree with pricing per token

• Why moving from Prague to the Bay Area felt essential for them as founders

Roman Gershman is Co-Founder & CTO of Dragonfly, the drop-in Redis replacement for heavy data workloads that has significant performance, cost, and scale benefits. Their open source dragonflydb has 28K stars on GitHub.

Dragonfly has raised $21M from investors including Quiet Capital and Redpoint.

In this episode, we dig into:

• The challenges with Redis

• The users that have really benefitted from Dragonfly (high scale + real-time needs - gaming, B2C)

• The benefits of being multi-threaded

• How they got some of their bigger users / customers like Twilio, SoFi, and Spotify

Lukas Schulte is Co-Founder & CEO of SDF Labs, the developer platform that scales SQL understanding across organizations, which was recently acquired by data transformation unicorn dbt Labs. In this episode, he's joined by Anders Swanson, Senior Developer Experience Advocate at dbt, to discuss the acquisition and future of data engineering.

In this episode, we dig into:

• How the acquisition happened, as well as the M&A process

• How dbt thinks about building capabilities internally vs. making acquisitions

• How the SDF platform will improve the lives of dbt users

• The most challenging parts about the integration

• What the future developer experience for data teams will be like

• A glimpse into the future of data engineering