Built by analytics engineers, for analytics engineers
Loomkindle grew out of a direct encounter with the pain of maintaining semantic consistency across a growing data stack — and the conviction that this problem was solvable without heroic manual effort.
Make clean, semantically consistent data the default state for every analytics team — not the reward for heroic manual effort.
Most analytics teams spend the majority of their engineering time not building new capabilities, but defending the correctness of data they already have. A column renames upstream. A join key shifts. Three dashboards now disagree on the same metric, and no one knows which one is right. The on-call engineer traces the breakage for four hours and finds a schema change from six weeks ago.
That cycle is not inevitable. It is a tooling problem. Loomkindle exists to solve it by making the semantic contract between upstream data producers and downstream business consumers explicit, versioned, and automatically enforced. When that contract exists, schema changes surface immediately as a proposed review rather than a silent breakage. Metric definitions are canonical, not tribal knowledge. The analytics team builds, instead of repairs.
From a quarter-close Slack message to a production-grade platform
Wei Tan spent three years as a senior analytics engineer at a Seattle-based B2B SaaS company, watching his team rebuild the same fragile ETL scripts from scratch every time their data warehouse schema changed. The pattern was always the same: the VP of Finance would fire off a panicked Slack message at quarter close, the dashboard numbers would not reconcile, and the analytics team would spend a week tracing the breakage back to a column rename that happened six weeks earlier and silently propagated downstream.
The root cause was never bad SQL. It was the absence of an authoritative semantic contract between upstream data producers and downstream business consumers. Finance, Product, and Operations each maintained their own mental model of what "ARR" or "active user" meant, and those models diverged silently over months. The warehouse held the raw facts; no layer held the agreed-upon meanings.
Wei built an internal dbt macro library and a YAML-based metric registry that linked column definitions to business glossary terms. Within two quarters the fragile-pipeline incidents dropped 70% and the analytics team reclaimed 20 hours per sprint. His manager asked why he had not shipped this as a product.
Loomkindle is that internal tooling rebuilt as a production-grade platform — with an agentic discovery layer, declarative transformation compiler, and semantic catalog API serving analytics teams at seed-stage companies. The founding team of four met across two Seattle-area analytics engineering roles and set out to make the governed semantic layer the default, not the exception.
Loomkindle is a seed-stage company. We are building with a close-knit team of four, shipping fast, and working directly with early analytics teams to make sure every feature earns its keep. We are not targeting large-org contracts yet — we are earning early-customer trust one semantic layer at a time.
At seed stage, proximity to early customers is the strategic advantage. Every new connector we build, every edge case in late-arriving records, every contract violation pattern we encounter comes directly from an analytics engineer who uses Loomkindle in a real production environment. That feedback loop is not something we plan to outsource to a sales team and an enterprise requirements process. We stay close to the people who use this tooling, and that proximity is exactly how we earn the right to build further.
The principles that shape every product decision
Engineer empathy
We build for the person on-call at 2 AM, not the executive watching the demo. Every alert carries remediation context. Every error message names the exact table and column that failed. We measure UX by time-to-resolution, not time-to-impressed.
Semantic precision
Every metric definition in Loomkindle is explicit, versioned, and owned by a named engineer. We do not accept "it depends on which dashboard you're looking at" as an answer. Ambiguity is a bug, and we treat it like one.
Incremental trust
We earn confidence with small, auditable changes rather than big-bang migrations. The agent proposes; the engineer approves. Compiled SQL goes to git before it touches the warehouse. No black-box transformations, ever.
Open by default
Compiled SQL, versioned contracts, and public catalog APIs that teams own — not locked inside a proprietary runtime. If Loomkindle disappeared tomorrow, your analytics stack would still work. We think that is the right way to build software that handles critical data infrastructure.
Four engineers who lived the problem
Loomkindle's founding team met across two Seattle-area analytics engineering roles. Each person joined because they had personally spent months maintaining the exact problem Loomkindle solves.
Wei Tan
CEO & Co-Founder
Naomi Osei
CTO & Co-Founder
Faye Lindqvist
Head of Product
Rohan Iyer
Head of Engineering
Ready to make clean data the default?
Get early access to Loomkindle and ship your first governed semantic layer in a day — not a quarter.