DigDeeper's answer

Streaming recommendations optimise for what a broad audience will keep listening to. That is the opposite of what a DJ or selector needs, which is often the record nobody has played yet. DigDeeper has no popularity bias in its ranking, results are ordered by sonic similarity alone. It also lets you sort an artist's or label's entire catalogue by similarity to a track you already like, so a back catalogue of several hundred releases can be assessed in minutes instead of played through release by release.

DigDeeper's answer

DigDeeper works from the audio signal, not from metadata. Most discovery tools infer similarity from tags, playlists, or what other listeners played next — so a track needs to be some kind of popular before it can be recommended. DigDeeper analyses how a track actually sounds, which means a release with a few plays is as findable as a chart record.

That solves a specific problem: finding records that hold together as a set. Tracks sharing a sonic character mix well, and DigDeeper surfaces them in minutes instead of weeks of digging, including releases almost nobody knows yet. For DJs playing niche music, that is the whole point.

Search also chains: every result can become the next reference track, so you steer towards a sound step by step rather than guessing better search terms.

DigDeeper's answer

DigDeeper was built by Antonio Haas, who DJs as TONEE, and Marc Becker one DJ, one engineer, both into house music, both tired of getting the same Discover Weekly suggestions every Monday.

The frustration was specific. When you are hunting one particular groove, streaming algorithms hand you safe averages, Bandcamp browsing is luck, and working through label catalogues, Discogs and YouTube is slow and noisy. Tag-based search does not help either, because the tag says nothing about how a record actually sounds.

So we built the opposite: a search engine that ignores trends, tags and popularity entirely and looks only at the track you drop in.

DigDeeper's answer

DJs and selectors working in electronic music, from club and radio DJs to serious record collectors and crate diggers. Alongside them, professionals who evaluate catalogues for a living: A&R and label staff, music supervisors matching a brief to a specific sound, and reissue and compilation curators. What they have in common is that they search for a sound rather than a title, and that tag-based search has already failed them.

DigDeeper's answer

The core of DigDeeper is a deep audio model. It listens to a reference track and converts it into a numerical fingerprint capturing rhythm, bass weight, drum patterns, harmonic texture and atmosphere. Metadata plays no part: genre labels, artist tags and play counts are ignored.

That fingerprint is matched against the catalogue and the closest results are returned with a similarity score from 0 to 100%. It works like a reverse music search engine, where the query is a track rather than text. Because matching is purely acoustic, a 1994 jungle B-side and a 2024 release can score 94% similar if they share the same texture and energy, even when their tags have nothing in common.

The web application around it is a fairly standard modern stack; the audio model and the similarity search are the interesting part.