• 

  WorkTalk Coach is a workplace English learning App for Chinese English Learners

• 

  There are four paths including: Job Seekers path, First Workplace Role path, Communite Better at Work path, and Future Manager path

• 

  Each path contains 6 units, each completion will unlock the next unit.

• 

  Each unit comprehensively covers a topic from many aspects

• 

  In Scenario Practice, one can virtually practice real world workplace interactions with different people where an AI will provide feedback.

• 

  Tone Rewriter will adapt the same message for different audiences with the help of AI

• 

  Phrase Bank stores can store everything important or worth looking back, including rewrites!

WorkTalk Coach: Elevating Professional Presence

Inspiration

For many non-native English speakers, especially Chinese professionals, the "Glass Ceiling" in global companies isn't just about technical skill; it is about the Communication Gap. We noticed that while many possess a high level of written proficiency, they struggle with the professional register and nuance required in high-stakes environments. We were inspired to build a tool that doesn't just teach "English," but teaches "Professionalism." We wanted to create a "safe harbor" where users could fail, iterate, and refine their executive presence without the pressure of a real boardroom.

What it does

WorkTalk Coach is a specialized web application that provides scenario-based business English training. It guides users through four career-critical paths: Job Preparation, New Hire Integration, Advanced Communication, and Management Leadership.

• Scenario Simulations: Users respond to realistic workplace prompts via voice recording.
• 5-Dimension AI Assessment: An integrated AI engine evaluates recordings based on Content, Vocabulary, Fluency, Grammar, and Professionalism.
• Phrase Upgrades: The app provides direct "Before → After" examples, showing users how to transform basic, literal English into high-impact business communication.
• Curated Knowledge: Each unit is paired with expert resources from institutions like Harvard Business Review and MIT.

How we built it

The project is built on a modern full-stack architecture:

• Frontend: React + TypeScript + Tailwind CSS for a responsive, accessible UI. We used shadcn/ui for a polished, professional aesthetic.
• Backend & Serverless: Supabase Edge Functions manage the integration between the client and AI services.
• AI Engine: We integrated Gemini 2.5 Flash for intelligent analysis and Speech-to-Text for transcription.
• Scoring Logic: We developed a complex prompt engineering system to ensure the AI acts as a "Senior Business Coach," providing nuanced feedback rather than just grammar checks.
• Mathematical Precision: To calculate the overall_score, we use a weighted average of the five dimensions: $$Score_{overall} = \frac{1}{n} \sum_{i=1}^{n} w_i \cdot s_i$$ where $s_i$ is the score for dimension $i$ and $w_i$ is the importance weight assigned to that competency.

Challenges we ran into

• The "Nuance" Problem: Traditional NLP models often miss "hedging language" or tone. We faced significant challenges in fine-tuning the AI to recognize when a phrase was grammatically correct but culturally blunt or unprofessional.
• Audio Handling: Processing raw audio streams in a browser environment across different devices required robust error handling and specialized MediaRecorder configurations.
• JSON Reliability: Getting a Large Language Model to consistently output structured JSON for complex, 5-dimension feedback required multiple iterations of prompt refinement and validation guards.

Accomplishments that we're proud of

• The "Phrase Upgrade" Feature: We succeeded in creating a feedback loop that feels like a real conversation with a mentor. Seeing the AI successfully suggest an "executive version" of a user's input is incredibly rewarding.
• Zero-Error Deployment: Maintaining a lint-clean, production-ready codebase throughout 19+ versions of rapid iteration.
• Holistic Curriculum: Designing 72+ unique practice tasks that cover the entire lifecycle of a professional career.

What we learned

• AI as a Coach, Not a Critic: We learned that for learners, encouragement is just as important as correction. We structured the AI feedback to always include "Strengths" and "Encouragement" to keep users motivated.
• State Management in Scenario-Based Apps: Managing the state of recording, transcribing, and assessing while keeping the UI responsive taught us a lot about asynchronous React patterns.
• Prompt Engineering: We discovered that the quality of AI feedback is 90% dependent on the persona defined in the system instructions.

What's next for WorkTalk Coach

• Real-time Mock Interviews: Moving from asynchronous recording to a live, conversational AI agent that can ask follow-up questions.
• Community Benchmarking: Allowing users to see how their "Professionalism Score" compares to industry standards.
• B2B Integration: Partnering with global HR departments to use WorkTalk Coach as an onboarding tool for international hires.

Built With

• edge-functions
• gemini-2.5-flash
• lucide-react
• mediarecorder-api
• medo
• medo-gateway-api
• postgresql
• radix-ui
• react
• shadcn/ui
• sonner
• speech
• supabase
• tailwind-css
• typescript
• vite
• web
• whisper-v3