05.11 Algorütm: Kuidas Modash 400 miljonit sisuloojat otsitavaks teeb?

Summary

The episode examines Modash, an influencer-marketing platform that helps brands find creators and manage collaborations. The company says it indexes more than 400 million public creator profiles with over 1,000 followers. It has around 90 employees and 2,500 customers. Most customers are brands, though other influencer platforms can also build products on top of Modash’s data.

What the platform does

Brands can search for creators by factors such as location, language, audience interests, and the subjects they actually cover. The platform also supports campaign tracking, checks whether published content meets campaign requirements, monitors competitors, and helps manage creator collaborations. Modash is adding broader brand-mention and sentiment monitoring.

Building searchable creator data

Modash analyzes creators’ public posts rather than relying only on profile descriptions. Its pipeline first filters out content unlikely to be useful, such as blank or blurry images. It then uses multimodal language models to describe content and infer its subject, intent, and likely audience. Text descriptions are converted into embeddings, or vectors, for search.

Data products and insights

Modash can map which brands work with which creators and how creators collaborate with one another. These insights can help brands understand competitors’ strategies and avoid overlapping with them. Not every data product covers the full 400-million-profile database; coverage depends on available resources.

Scale, architecture, and cost

According to the guest, the system processes large volumes of public social-media data and makes more than two billion requests per month. The team continually balances coverage, processing depth, speed, and cloud costs—for example, deciding how much of a video to analyze and which content warrants more expensive model processing.

The company uses a mix of roughly a dozen open-source models and commercial models, including tools from OpenAI, Google, and Mistral. The guest estimates that about 30% of development effort goes toward maintaining existing systems, with the rest split between improving products and building new ones.

Platform access and creator preferences

Modash says it collects public data in a manner comparable to a search engine, rather than secretly accessing private content. Changes to social platforms can still disrupt data pipelines, and the team expects to handle such changes regularly. Creators can request removal from the database.

Market and product choices

Modash chose to build a searchable data platform rather than a two-sided marketplace, which would require attracting and activating both creators and brands. Its customers are spread between the US and other markets. The company is currently focused on Western markets and is not yet pursuing Asia.

AI content and evaluation

Some customers want to filter out AI-generated creators, and Modash is working on that capability. More broadly, evaluating model output is difficult because labels such as creator location, content intent, or a profile’s “vibe” can mean different things to different customers.

People still review prompt changes and sample outputs to judge whether results have improved. Making subjective preferences measurable remains an open challenge.

Reviews, guides, or tutorials

None are featured. The episode is an interview and technical discussion.

Main speakers and sources

Hendry Sadrak, Modash co-founder and technology lead; hosts Priit Liivak and Erik Jõgi; and Martin Kapp, who joins as a guest host.