SOTA · AI Analysis Platform

Project overview项目概览

Turn model outputs into shared, inspectable evidence把模型输出转化为团队可以共同理解的证据

Machine-learning scientists learn from large datasets, build models, compare experiments and continuously revise their work. SOTA gives them one protected environment to visualize model performance, share results and develop better algorithms together.

机器学习科学家需要从海量数据中训练模型、比较实验，并持续修正结果。SOTA 将模型表现可视化、实验管理和团队协作组织在同一个受保护的环境中，让科学家能够共同理解结果并持续优化算法。

Company机构SEA AI Labs × NUS

Time时间Sep 2021 · 2021 年 9 月

Role角色User researcher and UX designer用户研究员与交互设计师

Outcome成果A secure analysis and collaboration platform supporting game and commerce algorithms.支持游戏、电商等算法团队的数据分析与协作平台。

01

Scattered workflow工作流分散

Data, jobs, experiments and visual outputs lived across separate tools.数据、任务、实验和可视化结果分散在不同工具中。

02

Hard to compare结果难以比较

Scientists needed flexible plots and tables to understand model differences.科学家需要灵活的图表与表格来理解模型差异。

03

Weak collaboration协作上下文缺失

Screenshots and links could not preserve experiment context or ownership.截图和链接无法完整保留实验上下文与协作关系。

01 · User research用户研究

Understand the full scientific workflow before designing charts在设计图表之前，先理解科学家的完整工作流

Visualization was only one part of the task. Research focused on how scientists formed questions, logged data, managed teams and projects, compared experiments and shared findings.可视化只是整体任务的一部分。研究首先理解科学家如何提出问题、记录数据、管理团队与项目、比较实验并分享结果。

Existing behavior现有行为

A model begins with a question, not a dashboard模型从问题开始，而不是从看板开始

We mapped how a scientist learns from historical data, identifies influential factors and predicts a new outcome. This reframed SOTA from a visualization tool into an end-to-end analysis workspace.我们梳理科学家如何从历史数据中学习、寻找影响因素并预测新结果。这让 SOTA 从单一可视化工具转变为端到端的分析工作台。

INPUTHistorical data and experiment logs历史数据与实验日志
ANALYSISFactors, trends and model comparisons因素、趋势与模型比较
DECISIONA prediction that can be explained and shared可解释、可分享的预测结果

Diagram showing a machine learning scientist learning from data, finding factors and predicting an outcome — **Workflow model**The scientist's reasoning became the foundation for the product structure.科学家的推理过程成为产品结构的基础。

SOTA user research plan — **Research plan**Research questions, participants, interview structure and synthesis.研究问题、参与者、访谈结构与分析方法。

Affinity mapping and notes from SOTA user interviews — **Interview synthesis**Needs were grouped by motivation, frequency and product value.按照动机、频率和产品价值聚类用户需求。

Primary user persona for the SOTA platform — **Primary persona**A shared reference for the scientist's goals, habits and constraints.统一团队对科学家目标、习惯与限制的理解。

Persona用户角色

Barry needs one place to compare, collaborate and learnBarry 需要一个可以比较、协作与学习的统一空间

The persona helped the team move beyond individual requests and align around a coherent platform: integrated visualization, flexible comparison and collaboration without losing privacy or ownership.用户角色帮助团队超越零散需求，围绕一个完整平台建立共识：整合可视化、灵活比较，同时保护数据隐私和实验归属。

SOTA user journey map across data logging, project management, experiments, tables and sharing — **User journey**Pain points were mapped across logging data, managing projects, analyzing experiments and sharing results.将数据记录、项目管理、实验分析和结果分享中的痛点放进同一条旅程。

SOTA functional modules for teams, projects, experiments, plots and tables — **MVP scope**The MVP balanced user needs, technical constraints and product value.MVP 在用户需求、技术限制与产品价值之间取得平衡。

02 · Design solution设计方案

Organize experiments around projects, evidence and people围绕项目、证据与协作关系组织实验

The product architecture connects team management, projects, experiment plots, table views and profiles while preserving a clear route from raw data to shared conclusions.产品架构连接团队管理、项目、实验图表、表格视图和个人资料，并保留从原始数据到共享结论的清晰路径。

SOTA information architecture showing teams, projects, experiments and analysis views — **Information architecture**A product map aligned the team before detailed interaction work began.在进入详细交互设计前，先用产品地图统一团队方向。

01

Project context first项目上下文优先

Experiments stay connected to teams, projects and owners.实验始终与团队、项目和负责人保持关联。

02

Flexible analysis灵活分析

Plots and tables support different comparison and inspection needs.图表与表格支持不同的比较和检查需求。

03

Shareable evidence可共享的证据

Notes, permissions and links preserve the meaning of a result.备注、权限和链接共同保留结果的含义。

Interaction systemDetailed states covered project lists, plots, tables, administration and profiles.详细交互覆盖项目列表、图表、表格、管理后台与个人资料。

Validate before visual polish视觉细化前先验证

Before moving into final visual design, we returned to users for the first validation round. Testing interaction logic early reduced rework and confirmed that the platform matched scientists' expectations.进入最终视觉设计前，我们先与用户进行第一轮验证。提前测试交互逻辑减少了返工，也确认平台整体符合科学家的工作预期。

03 · Product experience产品体验

A unified place to inspect model performance and work together在统一空间中查看模型表现并协同工作

The final visual design turns complex experiment data into a restrained workspace with clear project, plot and table views.最终视觉方案将复杂实验数据整理为克制、清晰的工作台，统一项目、图表与表格视图。

SOTA product walkthroughUse the controls to play使用控件播放演示

04 · Product iteration产品迭代

Keep users inside the product's learning loop让用户持续参与产品的学习与迭代

SOTA treated iteration as a shared planning activity. User feedback, product value and engineering resources were translated into a roadmap that was specific in the near term and flexible in the long term.SOTA 将迭代视为共同规划的过程。用户反馈、产品价值和工程资源被转化为近期具体、远期灵活的路线图。

User groups用户群组

Maintain immediate contact with active scientists.与活跃科学家保持即时联系。

Public request list公开需求列表

Let users record problems and requests when they occur.让用户随时记录遇到的问题和需求。

Visible roadmap可见的路线图

Invite users into the product iteration lifecycle.让用户参与产品迭代的完整生命周期。

SOTA story map showing MVP and later phases across the product journey — **Story mapping**PM, design, engineering and QA aligned MVP work with later platform phases.PM、设计、开发和 QA 通过故事地图统一 MVP 与后续阶段。

05 · Outcome项目成果

Build the platform around how scientists actually learn围绕科学家真实的学习方式构建平台

Privacy隐私

Protected internal model data and experiment results.保护内部模型数据与实验结果。

Clarity清晰

Made model performance easier to inspect and compare.让模型表现更容易查看与比较。

Collaboration协作

Connected teams, projects, experiments and shared evidence.连接团队、项目、实验与共享证据。

Iteration迭代

Created a repeatable feedback and planning loop.建立可持续的反馈与规划闭环。

The right dashboard does not begin with charts. It begins with the decisions people need to make together.真正有效的数据平台，不是从图表开始，而是从团队需要共同做出的判断开始。

SOTA taught me to treat visualization as part of a larger system of workflow, collaboration and trust.SOTA 让我认识到：可视化必须被放进工作流、协作和信任共同构成的系统中设计。