AI Engineer
← All speakers

Related work and projects

  • Little Morgan ↗

    Suo created a multilingual assistant using Vapi and MiniMax text-to-speech to answer questions about MiniMax. She sought feedback while refining its prompts and speech recognition.

    linkedin.com
  • Pencil × MiniMax integration ↗

    Suo highlighted the integration and described continued work with Pencil. The creative platform offers MiniMax models for video, images, chat, copy and reasoning; Pencil advocates matching each brief to a model sufficient for the job.

    linkedin.com
  • MiniMax H3 ↗

    MiniMax announced that Suo was scheduled to discuss H3 at ModCon. The product announcement describes released base weights that generate video and audio from text or visual references; audio references require an accompanying image or video. Separate hosted components process complex context and regenerate output at 2K.

    minimax.io · Source ↗

Bio, Work & Ideas

Morgan Suo

Conference affiliation: MiniMax

Morgan Suo is MiniMax’s head of US business development. Her work connects the company’s AI models with applications in creative production and voice interfaces.

Suo has highlighted the MiniMax integration in Pencil as a way to bring multimodal AI into agency workflows, including video editing and adaptation. Pencil offers MiniMax models for video, images, chat, copy, and reasoning. Suo has also represented MiniMax’s H3 video model. Its open-weight base generates video and audio from text alone or with visual references; audio references can also be used when accompanied by an image or video. Separate hosted components interpret complex inputs and regenerate output at 2K resolution.

Suo has built with the company’s voice technology herself. Her Little Morgan assistant uses Vapi and MiniMax text-to-speech to answer questions about MiniMax in multiple languages. She invited feedback while refining its prompts and speech recognition.

In discussions with customers, Suo has observed interest in APIs, fast deployment, and cost savings. She argues that technical teams should test and compare tools, choosing what solves the problem rather than relying on perceptions of political safety alone.

Read the topics behind these talks

1 conference talk

▶ Watch ↗

17:42

Is a Faster Model Actually Faster? Test the Whole Job — Morgan Suo, MiniMax

Read the full talk →

Key ideas

Scroll to read ↓

Morgan Suo uses a financial research task to connect model optimizations to the accuracy, cost and wait time of a report an analyst can actually use.

  • Measure time and cost to an acceptable report, including retrieval, retries, failed attempts and human repair.
    1:13 ↗
  • Sparse attention reduces work by selecting text blocks, so test whether important evidence remains findable at different positions in real documents.
    5:31 ↗
  • Lower precision can reduce data movement, but near-identical average scores can conceal changes on individual tasks; full-setup comparisons do not isolate their causes.
    7:59 ↗
  • Test serving load and reasoning effort against the deadline and quality needed for the actual job, rather than relying on throughput or first-token latency alone.
    11:47 ↗

References