This startup offers a build system and package manager that ensures consistent software dependencies for machine learning projects. Their software creates reproducible and distributable builds, enabling development teams to use the same tool versions for training, deploying, and monitoring models.
Funding
Funding not disclosed
Founders
Product
Problem
Building and testing software, especially for machine learning projects, often suffers from inconsistent dependencies across different environments. This leads to unreliable builds, difficulties in reproducing results, and challenges in distributing software components.
Solution
Tangram provides a build system and package manager that ensures fast and reliable software construction. It locks project dependencies, including compilers, interpreters, binaries, and libraries, within a lockfile. By breaking down tasks into independently cacheable steps, Tangram facilitates the reuse of work across machines and distributes workloads efficiently. The system is programmable with TypeScript, offering features like autocomplete and type checking for build configurations. Tangram utilizes a content-addressed storage system accessed through a virtual filesystem to minimize disk usage and network transfer.
Target Audience
Tangram is designed for software development teams, particularly those working on machine learning projects, who require a reliable and reproducible build system and package manager.
Features
- Dependency locking via a lockfile, capturing all compilers, interpreters, binaries, and libraries
- Task decomposition into small, independently cacheable steps for efficient reuse and distribution of work
- TypeScript programmability for build configurations, enabling autocomplete and type checking
- Content-addressed storage system with a virtual filesystem to minimize disk usage and network transfer
- Caching of incremental build artifacts for faster and more reliable builds
- Support for building reproducible environments that start instantly