What is SocialCoach
SocialCoach is an AI emotional intelligence coaching tool jointly open-sourced by The Hong Kong University of Science and Technology (Guangzhou) and Microsoft Research Asia. SocialCoach lets users fully rehearse difficult conversations with opinionated AI characters in a low-risk simulated environment, and after practice, provides sentence-by-sentence feedback based on their original words, scoring communication quality and conversation outcomes separately. SocialCoach includes 46 scenarios, supports custom scenarios and both Chinese and English, and recommends practice based on weaknesses using reinforcement learning.
SocialCoach's Main Features
-
Realistic Conversation Practice Partner: Fully rehearse difficult conversations such as asking for a raise or chasing a debt with AI characters that have goals and opinions.
-
Sentence-by-Sentence Intelligent Feedback: After practice, receive sentence-by-sentence feedback quoting the user's original words, with communication quality and conversation outcomes scored separately.
-
Personalized Recommendations: Based on reinforcement learning, recommends the most suitable practice according to your goals and proficiency in 34 social skills.
-
Scenario Library + Customization: Includes 46 scenarios covering 7 major domains, and supports creating your own scenarios by describing your real situation.
-
Bilingual in Chinese and English: Switch between Chinese and English anytime for the interface and practice content, and practice English communication along the way.
Follow on WeChat and reply "Open Source" to join theAI Open-Source Project Community Group
How to Use SocialCoach
- Online Trial: Open https://socialcoach.aurax.live，免注册直接开始对话练习。
- Local Deployment: Clone the GitHub repository →
pnpm install→ configure.env.local(fill in the LLM Provider and API Key) →pnpm dev, and visit localhost:3000. - Docker / One-Click Deployment: Use the project's built-in docker compose or one-click deployment on Vercel or ModelScope, suitable for providing services externally (for self-hosting in China, note to set up
LLM_OPENAI_TOKEN_PARAM=max_completion_tokenscompatible with domestic endpoints).
SocialCoach's Core Advantages
- Characters Have Opinions: AI characters come with their own goals and bottom lines, will genuinely push back and bring up old grievances, and after practice you can withstand the test of real scenarios.
- Targeted Feedback: Feedback quotes the user's original words sentence by sentence, and can score "communication quality" and "conversation outcomes" separately.
- Systematic Practice: Based on the CASEL educational framework and reinforcement learning scheduling, supported by 42 communication strategies + 30 real cases.
- Broad Scenario Coverage: 46 preset scenarios cover 7 major domains including the workplace, interpersonal relationships, and family, and support customizing real issues.
SocialCoach's Project Address
- Project Website：https://tianfuwang.tech/SocialCoach/
- GitHub Repository：https://github.com/GeminiLight/SocialCoach
Comparison of SocialCoach with Similar Competitors
| Dimension | SocialCoach | Yoodli |
|---|---|---|
| Product Positioning | AI emotional intelligence conversation coach: practice "difficult conversations" | AI speech/expression coach: practice "speaking more pleasantly," often called the "voice version of Grammarly" |
| Core Capabilities | 34 social skills (communication strategies, conflict handling, boundary setting, etc.) | Spoken expression delivery: speaking speed, verbal tics, clarity, eye contact, filler words |
| Conversation Method | Multi-turn conversations with "nitpicking" AI characters that have goals and bottom lines | Simulate an interviewer/client/audience, or upload a video for speech analysis |
| Feedback method | Comments quoting your exact words sentence by sentence, with communication quality and results scored separately | Real-time or post-meeting report: speech rate, pauses, filler words, eye contact, body language |
| Teaching system | CASEL five competency framework + knowledge base of 42 strategies/30 cases + reinforcement learning personalized recommendations | Feedback-driven practice, with a relatively weak personalized recommendation system |
| Covered scenarios | 46 scenarios across 7 major domains, with custom support | Scenarios leaning toward "public expression," such as interviews, speeches, meeting reports, and sales conversations |
| Technical form | Open source (Apache 2.0), self-hostable, commercially usable, bring your own API Key | Closed-source commercial product, web + App |
SocialCoach application scenarios
-
Workplace expression: asking your boss for a raise, refusing unreasonable overtime, and how to respond when questioned in a meeting.
-
Boundary setting: getting a friend to pay back money, setting clear rules with a roommate, and refusing requests from people who keep pushing their luck.
-
Conflict repair: discussing household chores with a partner, talking with parents about career choices, and repairing a relationship after an argument.
-
Social connection: welcoming a new colleague, supporting a friend who is feeling down, and naturally joining a conversation with strangers.
-
Dual practice in language + communication: switch to English mode to practice English expression and communication strategies at the same time; customize scenarios to rehearse interviews, apologies, and any other real challenges.