Ethan Steininger

Former lead of MongoDB's Search Team, Ethan noticed the most common problem customers faced was building indexing and search infrastructure on their S3 buckets. Mixpeek was born.

Recent Posts

🚀 The Rise of the Dataset Engineer

Why the future of AI isn’t about bigger models — it’s about better data.

May 12, 20255 min read

Understanding Late Interaction Models in Multimodal Retrieval

Late interaction models enable precise retrieval from multimodal data like PDFs and images by comparing query tokens with token or patch embeddings—ideal for RAG, search, and document understanding.

May 8, 20254 min read

Video Segmentation: Unlocking Structure for Search and Analytics

Segmentation turns raw video into searchable chunks—objects, actions, scenes—boosting precision in multimodal search. It bridges unstructured content with structured queries like “man falling in warehouse,” enabling faster, more accurate retrieval across large datasets.

April 22, 202511 min read

AI-Powered Content Recommendations for AdTech

Even with data-driven targeting, most ads still miss. Contextual AI changes that—boosting relevance, clicks, and ROI without cookies.

April 13, 20252 min read

Contextual Advertising After the Death of Cookies

As Google phases out third-party cookies, advertisers face declining performance from behavioral targeting. Learn how Contextual AI offers a privacy-safe, high-precision alternative.

April 8, 20255 min read

🎯 Multimodal Monday #2 — From Tiny VLMs to 10M‑Token Titans

📢 Quick Take (TL;DR) * Major multimodal model releases: Meta unveiled Llama 4 Scout & Maverick – open Mixture-of-Experts models with native text+image (and even video/audio) support – and Microsoft introduced Phi-4-Multimodal, a compact 3.8B-param model integrating vision, text, and spee (Today is the start of a new era of natively multimodal AI… | AI at Meta | 190 comments) ([2503.01743] Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs)7】. Bot

April 7, 20257 min read

Turning Frames into DataFrames: AI-Powered Video Analytics

By applying the classic group_by pattern to structured video data at index time, you can turn raw frames into searchable, analyzable DataFrames aligned with how your users explore footage.

April 6, 20255 min read

🧠 Multimodal Monday #1 - State of the Stack

Researchers introducing new methods to replace embeddings with discrete IDs for faster cross-modal search

April 2, 20255 min read

Semantic Video Chunking: Scene Detection

Intelligent video chunking using scene detection and vector embeddings. This tutorial covers how to break down videos into semantic scenes, generate embeddings, and enable powerful semantic search capabilities.

February 5, 20256 min read

AI Video Tagging With Dynamic Taxonomies

AI video tagging used to mean manual review and basic object detection. With multimodal models and dynamic taxonomies, you can now automatically detect brand moments, inappropriate content, actions, moods and trending content at scale.

December 30, 20244 min read

Building your own AI-Powered Media Asset Management System

This guide will walk developers through building a modern Media Asset Management (MAM) system with semantic search capabilities using Mixpeek's infrastructure.

December 29, 20244 min read

Reverse Image Search with CLIP and MongoDB

AI-powered image discovery app using Mixpeek's multimodal SDK and MongoDB's $vectorSearch. Features deep learning, vector embeddings, and KNN search for advanced visual content management.

December 13, 20244 min read

Build a Model Context Protocol (MCP) Server on S3 using Lambda, Temporal, Ray, and Qdrant

Build a scalable MCP pipeline on S3 using AWS Lambda, Temporal, Ray, and Qdrant to process and index unstructured data like video, audio, and PDFs for real-time AI search and retrieval.

December 4, 20242 min read

Advanced Video Understanding: Mixpeek Embed and Weaviate KNN for Multimodal AI

This article demonstrates how to build a reverse video search system using Mixpeek for video processing and embedding, and Weaviate as a vector database, enabling both video and text queries to find relevant video segments through semantic similarity.

November 1, 20245 min read

NVIDIA Cosmos: The Makings of a World Foundation Model

World foundation models are neural networks that simulate real-world environments and predict accurate outcomes based on text, image, or video input.

November 1, 20244 min read