中文Talk to us

Feiliu · Multimodal recommendation engine

Let quality content meet the right audience, faster.

Feiliu turns our in-house multimodal recommendation engine into a product you can integrate, configure and extend — from recall and ranking to risk control and experimentation.

Talk to usExplore the product
pipeline stages, one governed flow
5
capability pillars, usable on their own
4
new platforms on the roadmap: iOS, Android, H5
3
to multi-platform launch
6–12mo

Capabilities

The full recommendation stack, as a service.

Use the whole pipeline or only the modules you need. Each one is configurable from the console and available through our SDK and API.

  • Recall

    Multi-strategy candidate retrieval built on vector search and multimodal signals from text, images and video.

  • Ranking

    Pre-ranking, ranking and re-ranking with diversity and exploration controls, driven by online features and real-time feedback.

  • Risk control

    Text and image review, anti-abuse strategies and configurable compliance modules that follow your own content standards.

  • Experimentation

    A built-in A/B platform with consistent tracking across platforms, so every change is measured before it ships.

How Feiliu ranks a feed

From the whole catalogue to one feed, in five stages.

Every request flows through the same governed pipeline. Each stage narrows the candidates and every decision leaves a trace.

Keep scrolling to step through the five stages.

  1. Stage 1 of 5: Recall

    Vector retrieval and multiple recall strategies pull candidates from the full catalogue, combining text, image and video signals.

    • Vector retrieval
    • Multi-strategy recall
    • Text · image · video

  2. Stage 2 of 5: Pre-rank

    A lightweight model trims the candidate set quickly, so heavier models only score what matters.

    • Lightweight model
    • Fast trimming
    • Saves heavy-model compute

  3. Stage 3 of 5: Rank

    Multimodal and sequence models score each item against online features and real-time feedback.

    • Multimodal models
    • Sequence models
    • Online features · real-time feedback

  4. Stage 4 of 5: Re-rank

    Diversity, exploration, frequency control and business weights shape the final order of the feed.

    • Diversity and exploration
    • Frequency control
    • Business weights

  5. Stage 5 of 5: Risk control

    Text and image review, anti-abuse strategies and your own content standards check every item before it is served.

    • Text and image review
    • Anti-abuse
    • Your content standards

Why Feiliu

Built for results you can explain.

We believe recommendation is worth it only when it is both effective and controllable: lifting clicks and conversion today while protecting long-term retention, your creator ecosystem and compliance.

  1. Multimodal engine

    Text, image and video signals fused in one model, with online features and a real-time feedback loop.

  2. Complete tooling

    Audience targeting, allow and block lists, a weight panel, frequency and diversity controls, and an A/B platform.

  3. Explainable and controllable

    Reason tags, strategy visualisation and configurable compliance modules — every decision can be traced and rolled back.

  4. Strong backing

    An independent company backed by strategic industry partners, with mature capability in infrastructure, model iteration, delivery methodology and team growth.

Typical scenarios

Wherever content needs to find its audience.

  • Content platform feeds

    Personalised home feeds and discovery for articles, images, short video and live content.

  • E-commerce and advertising

    Product and ad recommendations that balance relevance, conversion and business rules.

  • Vertical communities

    Feeds for focused communities that grow creators and keep discussions healthy.

  • Brand-owned channels

    Content distribution inside a brand’s own apps and private channels.

  • Solutions →

Where we are

Proven in a mini-program. Going multi-platform next.

The full loop already runs in our mini-program demo app, built by a core team with hands-on multi-platform experience. Over the next 6–12 months we will bring Feiliu to iOS, Android and H5, with the same tracking and experimentation on every platform.

  1. Mini-programFull loop live now
  2. iOSWithin 6–12 months
  3. AndroidWithin 6–12 months
  4. H5Within 6–12 months

Compared with building in-house

Better value than building it yourself.

Compared with assembling a recommendation team from scratch, Feiliu gets you live faster, keeps results under control and brings compliance governance with it.

  1. Faster to launch

    Start from a working pipeline, SDK and console instead of hiring and building everything from zero.

  2. Results under control

    A/B experiments, reason tags and versioned rollback are built in, so every change is measured and reversible.

  3. Compliance built in

    Review tools, your own content standards and a strategy audit trail come with the product, not as a separate project.

Bring better recommendations to your platform.

Tell us about your content and audience. We will propose a proof of concept scoped to your goals.

Talk to usSee the architecture