Enterprises can now train custom AI models from production workflows — no ML team required

Every interaction with an enterprise AI application, from queries to corrections, generates valuable training data. However, most organizations are not capturing this data, letting it go to waste. San Francisco-based Empromptu AI aims to change that with the launch of Alchemy Models, a platform that automatically captures and validates outputs from subject matter experts, routing them back into a fine-tuning pipeline to improve the model over time.

Enterprises own the resulting model weights outright, giving them control and customization capabilities. Alchemy Models sits in a different territory from both Retrieval-Augmented Generation (RAG) and traditional fine-tuning. While RAG retrieves external context at inference time without modifying model weights, traditional fine-tuning changes weights but requires separately assembled labeled datasets and a dedicated ML pipeline.

Alchemy does the latter continuously, using the enterprise application itself as the data source. Companies adopting foundation model APIs face significant constraints, including scaling inference costs, lack of model ownership, and limited customization capabilities. Empromptu CEO Shanea Leven notes that these constraints are widely felt but rarely addressed.

"Every customer, everybody that I talk to, is like, how am I not going to get disrupted? How am I going to protect my business? And they just don't see the path," Leven told VentureBeat in an exclusive interview.

The Alchemy platform builds a model from a running application by generating and cleaning training data through Empromptu's Golden Data Pipelines infrastructure. The mechanism runs in two stages: before an app is built, enterprise data is cleaned, extracted, and enriched, and once it's running, every output it generates goes back through the pipeline, where subject matter experts review and correct it. The validated output becomes the training data for the next fine-tuning run.

The resulting fine-tuned models are what Empromptu calls Expert Nano Models: small, task-specific models optimized for a particular workflow rather than general-purpose reasoning. The platform is model-agnostic, supporting Llama, Qwen, and other base models. One early user, behavioral health company Ascent Autism, used Alchemy to automate session documentation and parent communication, cutting session documentation time by up to 87%.

Leven positions Alchemy as a third architectural choice for enterprises, combining the ongoing improvement of fine-tuning with the operational simplicity of building inside a managed platform. "Having that data moat is the most valuable currency," Leven said. With Alchemy, enterprises can now train custom AI models from production workflows without requiring a dedicated ML team, giving them a competitive edge in the market.