ethan - about

Ethan Steininger

I build things. Right now that's Mixpeek, which makes it easier to search through video, images, and other messy data. Before that, I spent a few years at MongoDB working on search. I also tinker with amux, which lets AI agents work together.

I grew up in Maryland in a Jewish family. My grandfather survived the Holocaust but never really talked about it. That shaped a lot of how I see the world. I write about some of that under Articles.

When I'm not behind a computer, I'm usually wandering around NYC or somewhere on the road in a Ford Transit I converted myself, with a paddleboard on the roof and my Australian Shepherd, Dany, riding shotgun.

Shuffle: Sort by: Order:

Loading clusters...

Namespace Manifest

Give this manifest to your AI agent. It will set up image embedding, visual clustering with LLM-generated labels, and scheduled re-clustering on your own Mixpeek namespace.

version: '1.0'
namespaces:
- name: my-photo-album
  description: Photo album with image embeddings and visual clustering
  feature_extractors:
  - name: image_extractor
    version: v1

collections:
- name: album-collection
  namespace: my-photo-album
  description: Photos with image embeddings for clustering and search
  source:
    type: bucket
    bucket: album-photos
  feature_extractor:
    name: image_extractor
    version: v1
    input_mappings:
      image: image
  enabled: true

buckets:
- name: album-photos
  namespace: my-photo-album
  description: Source photos
  schema:
    properties: {}

clusters:
- name: album-visual-clusters
  namespace: my-photo-album
  source_collections:
  - album-collection
  algorithm:
    name: hdbscan
    parameters: {}
  output_mode: enrich_source
  cluster_type: vector
  llm_labeling:
    enabled: true
    provider: google
    model_name: gemini-2.5-flash-lite
    include_summary: true
    include_keywords: true
    sample_selection_strategy: nearest
    use_embedding_dedup: true
    embedding_similarity_threshold: 0.8

triggers:
- name: reprocess-collection
  namespace: my-photo-album
  description: Re-process collection every 6h for new uploads
  action_type: collection_trigger
  action_config:
    collection: album-collection
  trigger_type: cron
  schedule_config:
    cron_expression: '0 */6 * * *'
    timezone: America/New_York
  status: active

- name: recluster-daily
  namespace: my-photo-album
  description: Daily re-cluster at 6am
  action_type: cluster
  action_config:
    cluster: album-visual-clusters
  trigger_type: cron
  schedule_config:
    cron_expression: '0 6 * * *'
    timezone: America/New_York
  status: active

Apply with: POST /v1/manifest/apply — Mixpeek docs