Qwak

Qwak, whose platform is now presented as JFrog ML (formerly Qwak), is an end-to-end MLOps platform that lets data science and ML engineering teams build, train, deploy, manage and monitor machine learning and AI workloads in one place, covering classical ML as well as generative AI and large language models. The core product spans model building and a central model registry, one-click training and fine-tuning on CPU or GPU machines, deployment as real-time API endpoints, batch inference jobs or streaming models connected to Kafka with multi-version deployments, and production monitoring for model performance, data anomalies and input-distribution drift with alerting integrations such as Slack and PagerDuty.

Alongside MLOps, the platform provides an LLMOps layer — a prompt registry with versioning and a prompt playground, a model library for one-click deployment of optimized open-source models such as Llama 3 and Mistral 7b on the customer's cloud or Qwak's, visual construction of multi-step LLM workflows with shadow prompt deployments, and request tracing and monitoring of prompt content, inference calls and latency. A feature store component manages the full feature lifecycle, ingests and transforms data from warehouses and other sources through custom data pipelines, and stores embedding vectors at scale for vector search, recommendation engines and RAG pipelines that feed LLM applications.

The company positions the platform as a way to remove the bottleneck between engineering and data science teams and to reduce the infrastructure work required to move models from research into production, with a single environment shared by ML engineers, data scientists, product managers and other AI practitioners.

Founding story

The founding team met through prior roles at large technology companies: Alon Lev (CEO) was previously VP Data at Payoneer; Yuval Fernbach (CTO) was a machine learning specialist at Amazon Web Services and formerly CTO in the IDF's Mamram unit; Lior Penso (COO) worked at ironSource and AWS; and Ran Romano (VP Engineering) led machine learning engineering at Wix. Lev left Payoneer during the first Covid wave and recruited Fernbach, his former IDF commander, from Amazon; Fernbach brought in Penso, and Romano joined during the ideation phase. The founders said they started the company out of personal need, having repeatedly seen organizations discuss machine learning and decision-support models without the human or product infrastructure to put them into production.

Business model

Qwak sells a fully managed, end-to-end ML platform to companies rather than requiring them to assemble open-source tools or build in-house infrastructure; the CEO described a pay-as-you-go model that consolidates feature store, model registry, deployment, monitoring and pipeline orchestration with a single vendor.

The company describes a pay-as-you-go model for its fully managed, end-to-end ML platform.

Traction

Named customers include NetApp, Lightricks, Yotpo, JLL, Guesty, OpenWeb and Skyline AI. The company reported 10x year-over-year revenue growth as of March 2023 and said the Series A1 doubled its valuation. Headcount grew from 20 in February 2022 to 40 by April 2024, and total funding reached $27 million. A published customer example describes Guesty building a RAG-based support chatbot on the platform, with usage of the new chat app rising from 5.46% to 15.78%.

Latest developments

The platform is now marketed as JFrog ML (formerly Qwak), with an expanded LLMOps suite covering prompt management, a model library in private preview for one-click deployment of optimized open-source LLMs such as Llama 3 and Mistral 7b, LLM workflow building, tracing and monitoring, plus a vector store for RAG use cases. In April 2024 the company reported 40 employees and $27 million raised to date and placed second in the Globes most promising startups ranking.