Self-Serve Data Service

Strengthen Sahara’s business model and streamline task management by transitioning to user-initiated tasks.

#Sahara AI  #Task Creation  #English Platform

Scope
Let requester customize tasks based on their own requirements, so they can independently manage data service tasks, including uploading annotated data, setting quality requirements, defining budgets, tracking progress, and retrieving results.
Basic Info
Target User
Requester
Time
Jul 2025
My Role
Product Designer
Key Steps of a Task
View More
Keyframes
Use Case 1: As a requester, I need to choose a decent template for my labeling projects , then start to add task details
Preview Template
Use Case 2: As a requester, I need to finish in total 6 steps settings, they are setting basic info, uploading raw datas, designing label interface, writing instructions, set quality threshold, and adding a reward pool.
Set Basic Info
Upload Raw Data
Design Labeling Interface
Write Instructions
New Sample
Preview Samples
Set Quality Threshold
Add a Reward Pool
Submit for Review
Detail Page, Tab is Dashboard.
Tab = Data
Tab = Settings
Use Case 3: As a requester, I need to set a time range to activate the task.
Set the time range
Use Case 4: As a requester, I want to preview some completed datapoints, then distribute the reward pool.
Preview a labeled datapoint
Double confirm with the requester before distributing the final reward.
Once the reward pool is distributed, requester can download the labeled data.
Design Research
Main Competitors
SCALE AI
Toloka
Labelbox
Label Studio
View Analysis Process
Challenge 1: How to reduce the task-creation cost?‍
- Guidebook: Describe the whole process in a guide document.
- Consistency: Apply a left-and-right structure in each setting step. User can input content on the left, and preview the final look on the right in real time.
- Reusability: keep reusable functions consistent in design
- Quick editing: Pay attention on the UX of secondary editing after creation because most users won’t have enough materials ready for the first time.  
Challenge 2: How do we organize a user friendly task detail page? What we can learn from main competitors?
Scale A
  1. No main project detail page, user can visit each section through the menu in side bar
  2. Emphasize the management of batch progress
  3. Show key project Info
Labelbox
  1. Use tabs to organize all sections
  2. It's not user-friendly to configure the task, hard to get the overall process
Toloka
  1. Use tabs to organize all sections
  2. It’s easy to manage all tasks and monitor the progress.
Label Studio
  1. Focus on data management
  2. Place task basic info in secondary page, but it's easy to edit any settings.
Opportunities
  • The cost of task configuration is high. Secondary editing scenario is necessary
  • Item B
  • Reduce the cognitive cost of task creation. Optimize the scalability of single components, like labeling interface, instructions, data management.
Project Prioroty

After design delivery, the project priority was lowered and marked as pending.

Reason 1: Strategic adjustment

The company has discontinued the AI Marketplace platform. As a result, the project’s main C-end users — independent AI developers who used DSP data to train or sell models — no longer exist.

Reason 2: Limited B-end value

Although the feature could improve communication efficiency with enterprise clients, the financial return is low, and the current client scale can still be handled manually.

Overall Reflection

After the project concluded, the most striking impression was its complexity. This complexity mainly stemmed from numerous critical process nodes and specialized terminology (such as Instruction and Batch).

In such intricate business scenarios, product designers must strike a balance between understanding, prioritization, and structured expression.

‍Product Level‍

  • Clarify business scenarios and priorities: Identify which features are essential, which are optional, and which can be postponed in Phase 1.
  • Define key decision-makers: The product involves multiple connected platforms (e.g., the operations dashboard, AI Marketplace, and Data Service List).
  • Clear boundaries between upstream and downstream interactions are crucial for efficient collaboration and output quality.

‍Design Level‍

  • Maintain a holistic perspective: Consider each functional module’s adaptability and consistency across different contexts to reduce user effort.
  • Simplify interactions: Avoid unnecessary complexity; keep tasks straightforward so users can focus on content input rather than interface logic.