LSRW Communication Assessment Platform

Educational institutions needed a consistent way to assess communication skills at scale. The platform delivers LSRW tests as interactive modules, grades responses with AI where that is reliable, and routes subjective work to educators for review, all with live reporting.

Manual LSRW evaluation is slow and inconsistent across evaluators, and speaking or storytelling tasks are hard to grade objectively. Institutions running hundreds of students per session cannot wait days for results.

• 20+ testing modules covering dictation, comprehension, storytelling, response selection and speaking tasks
• AI grading for both spoken and written responses
• Hybrid review support so educators can override or grade subjective tasks
• Test orchestration for large sessions with real-time reporting for educators

• Built the platform across Next.js frontend, Node.js backend and MongoDB
• Designed the modular test engine so new module types could be added without touching existing ones
• Integrated OpenAI-based grading for speech and written responses and the hybrid review workflow around it
• Built the orchestration and analytics pipelines behind real-time reporting

1. 01

   Student responds

   A spoken or written response is captured for a module task.

2. 02

   AI grading

   The response is graded through the OpenAI API.

3. 03

   Review routing

   Objective tasks are finalized automatically; subjective tasks enter the hybrid review queue.

4. 04

   Reporting

   Grades feed analytics pipelines and real-time educator reports.

• Modular test engine with a shared module contract for 20+ task types
• Grading pipeline wrapping the OpenAI API with a hybrid review state for subjective tasks
• Orchestration and analytics pipelines sized for 1,000+ assessments per session

1. 01

   Trusting AI on subjective tasks

   Challenge

   Storytelling and speaking tasks do not have a single correct answer, so fully automated grading risked unfair results.

   Solution

   Automated the objective tasks and built a hybrid review path for subjective ones, keeping educators in control while still removing 60% of manual evaluation time.

2. 02

   Reporting during large sessions

   Challenge

   Institutions ran 1,000+ assessments in a single session and needed results while the session was still running.

   Solution

   Built orchestration and analytics pipelines that process assessments as they complete and surface real-time reports to educators.

• Hybrid review keeps a human decision on subjective tasks rather than trusting model output blindly
• Module contract lets each task type define how it is graded, so grading logic stays explicit and testable

• Next.js
• Node.js
• MongoDB
• OpenAI API

Adopted by 3+ educational institutions. Manual evaluation time dropped by 60% and sessions of 1,000+ assessments report in real time.

No public link is available for this project.