LinkedIn Open-Sources FeatureFu, A Toolkit For Building Machine Learning Models
Frederic Lardinois / TechCrunch :
Context & Ripple Effects
FeatureFu lands eight months after Facebook's pledge to start building things in the open with its own deep-learning release, making 2015 the year large consumer-web companies began treating their internal ML tooling as a public asset rather than a trade secret. For LinkedIn, the toolkit exposes the feature-engineering machinery behind its own models.
The move also sits at the front end of a pattern that kept repeating: Microsoft packaged comparable capabilities into paid Azure machine learning services two years later, Facebook shipped its DLRM recommendation model for benchmarking in 2019, and Uber released its code-free Ludwig toolbox the same year — while LinkedIn itself eventually moved up the stack to OpenAI-powered product features.
First-order effects
- Outside developers get free access to the same feature-building toolkit LinkedIn runs internally, lowering the cost of assembling ML models for teams without dedicated infrastructure.
Second-order effects
- Open-sourcing tooling becomes a competitive expectation among platform companies — each release pressures peers to match it, as Facebook, Uber, and Microsoft's successive ML releases show.
Third-order effects
- If the pattern holds, the model-building toolkit layer gets commoditized by free corporate releases, shifting competitive advantage toward what stays closed: proprietary data, distribution, and application-layer products like LinkedIn's later AI features.
The trend: Consumer-web platforms are normalizing open-sourcing their internal machine learning toolkits, commoditizing the tools layer while keeping data and applications proprietary.