LastMile AI

LastMile AI is a platform aimed at assisting software developers in incorporating GenAI models into their applications. It was founded by former Meta product engineering team members Sarmad Qadri, Andrew Hoh, and Suyog Sonwalkar.

This platform enables customers to develop GenAI applications using text and image generation models from various sources, including open- and closed-source providers. Developers have the ability to customize these models using their data and seamlessly integrate them into their new or existing applications, products, and services.

The platform offers three key products: AI Workbooks, AI Workflows, and AIConfig.

LastMile's AI Workbooks offers a notebook-inspired environment for rapidly prototyping and building apps with AI models for text, image, and audio. Features include access to multiple AI models like GPT-4 and Stable Diffusion, parameterization for creating reusable AI templates and collaboration tools to share with an organization. AI Workflows allows the automation of tasks using GenAI technology. It supports integration with various data sources like Google Drive and CRMs, transformation of data using AI, and connection to different outputs like Slack. Lastly, AIConfig, available on GitHub, is a configuration-based framework aimed at constructing GenAI applications. It focuses on organizing prompts, models, and settings using JSON-configurable structures for better version control and swift prototyping.

The company's platform is designed for the entire lifecycle of AI app development, allowing for experimentation, prompt engineering, evaluation, deployment, and integration of AI models into apps and workflows. This approach aims to reduce the barriers to entry for non-machine-learning software engineers in developing AI-powered applications.

In November 2024, the company introduced AutoEval, a fine-tuning platform for evaluator models. AutoEval provides customizable evaluator models for testing AI applications, featuring synthetic data labeling and alBERTa for metrics like relevance, equivalence, and toxicity, with sub-300ms CPU inference.

The platform is offered through a SaaS subscription model, which includes a Free package for individual users, a Growth package for USD 50 per month per seat, and custom pricing for enterprises (March 2024).