RecipeBook by Shofo

Product Introduction

1. Definition: RecipeBook by Shofo is a self-service video data marketplace and semantic search platform for machine learning (ML) and artificial intelligence (AI) model training. It provides programmatic access to a catalog of over 25 million video clips.
2. Core Value Proposition: It exists to eliminate the traditional friction in sourcing high-quality, specific video datasets for computer vision projects. Its core value is enabling AI researchers and developers to semantically search a massive video database, iteratively refine results using active learning, and instantly purchase the exact curated datasets they need without lengthy sales processes.

Main Features

1. Semantic Video Clip Search: The platform allows users to search the 25M+ clip repository using natural language queries. Unlike keyword-only systems, it uses deep learning models to understand the semantic content of both the query and the video, returning contextually relevant results for terms like "person opening a door" or "vehicle making a left turn."
2. Iterative Ranking with Active Learning (Train a Probe): This is the core technical workflow. Users can rate (thumbs up/down) a small subset of initial results. RecipeBook then uses these human votes to train a lightweight probe model on-the-fly. This model re-ranks the entire catalog based on the user's specific preferences, dramatically improving relevance without manual review of millions of clips.
3. Self-Service Data Marketplace: The entire process—search, iteration, and purchase—is automated. Users can define their dataset, get a price based on compute time (priced at $3 per hour for the training probe), and download their curated clips without any contact forms, sales calls, or custom negotiations.

Problems Solved

1. Pain Point: It solves the expensive, slow, and inefficient process of acquiring labeled video data for AI training. Traditional methods involve sourcing raw footage, manual filtering, and annotation, which is time-consuming and costly, or dealing with inflexible, enterprise-sales-driven data vendors.
2. Target Audience: Primary personas include Computer Vision Engineers, ML/AI Researchers, Autonomous Vehicle Teams, Robotics Companies, and Video Analytics Startups. It's tailored for technical practitioners who need specific visual data (e.g., specific actions, environments, or objects) to train or fine-tune models.
3. Use Cases: Essential for scenarios like: training a pedestrian behavior prediction model for autonomous driving (needing clips of "people jaywalking"), developing a retail analytics system (requiring video of "customers picking up products from a shelf"), or creating a sports AI (needing data for "tennis player serving").

Unique Advantages

1. Differentiation: Unlike static data marketplaces or raw video archives, RecipeBook is an interactive, intelligence-powered platform. It competes by offering dynamic curation powered by user feedback, whereas competitors typically offer pre-packaged, generic datasets or require full custom curation projects.
2. Key Innovation: The "search, rate, re-rank" loop powered by on-demand probe training. This application of active learning within a commercial data platform is unique. It allows the system to adapt to highly niche user needs in minutes, transforming a passive database into an interactive data discovery and curation engine.

Frequently Asked Questions (FAQ)

1. What is RecipeBook by Shofo used for? RecipeBook is used for sourcing and curating targeted video datasets to train machine learning models for computer vision applications, such as autonomous systems, activity recognition, and video understanding AI.
2. How does RecipeBook's semantic search work? It uses deep neural networks to encode both your text query and the video content into a shared vector space. It then finds video clips whose vector representations are semantically closest to your query, going beyond simple keyword or tag matching.
3. What does "train a probe" mean on RecipeBook? Training a probe refers to the platform using your ratings on a small sample of videos to quickly train a lightweight machine learning model. This model learns your preference pattern and re-orders the entire video catalog to surface the clips most relevant to your specific criteria.
4. How much does RecipeBook by Shofo cost? The platform operates on a transparent compute-time model. The active learning re-ranking process ("training a probe") is billed at $3 per hour. You pay for the curation time, then purchase the specific clips you select for your final dataset.
5. What kind of video data is available on RecipeBook? The platform provides access to a vast and diverse catalog of over 25 million video clips, suitable for training robust AI models across multiple domains like mobility, retail, security, and human activity analysis.