AI and ML Developer Tools in the Y Combinator W24 Batch

Featuring 22 Observability, Model Training and Deployment Platforms and Other Specialized Developer Tools

The selection of early stage companies chosen for each Y Combinator batch is an interesting indicator on the growth vectors in the software industry. And this year, more specifically, in the AI and Machine Learning space. This is a post looking at 22 developer tools that have been selected in the W24 batch of almost 250 companies, with representatives from three categories:

• Observability tools - aimed at providing more precise, comprehensive and coherent explanations of what is happening within AI models and pipelines

• Model training and deployment platforms - meant to speed up the entire AI development process

• Specialized AI developer tools - which target a specific part of the AI value chain to increase its performance, from data curation, to security and QA

Observability

Observability tools are meant to provide visibility into AI models' operations and performance in order to ensure that they work according to the desired performance standards and product objectives. They enable developers and data scientists to monitor, debug, and optimize AI models by tracking metrics such as prediction accuracy, model drift, and computing resource usage. AI observability platforms can identify performance bottlenecks, unexpected model behavior, and compliance issues, as well as recommend steps to fix or address problems.

The companies in the W24 batch in the observability segment are:

• Guide Labs, building interpretable foundation models

• Vectorview, developing custom benchmarks and evaluation agents

• OneGrep, an AI-based observability copilot

• Reprompt, focusing on hallucinations monitoring

• Rags, an open source observability solution

• Phospho, providing observability for text analytics

• Ragas is a platform for evaluation of LLM applications that comes with its own open source benchmarking standards. The tool offers model-graded evaluation and testing techniques, automated synthesis of test data points, explainable metrics and adversarial testing. Ragas is open source and accepts contributions on GitHub. 

• Guide Labs builds interpretable foundation models that can explain their reasoning and are easier to align. They do that by providing human-understandable explanations for their outputs, as well as which parts of the input prompt are more important for each part of the generated output and which training data led to a specific generated output. Because the models can explain their outputs, they are easier to debug, steer, and align, especially for those verticals where safety and accuracy are important, such as finance, healthcare, insurance.