How many times have you been sitting around waiting for someone to review your code? Ever wish you could ping your boss at 2am? Well we did, and unfortunately, our boss didn't like that..... Thus was born Mentral AI! Mentral AI aims to solve this issue by 'cloning' any peer, mentor, or role model constantly evolving and learning as per their code, communication, and feedback styles. This gives you a 24/7 pair program, constantly looking over your code providing valuable feedback in the same manner you are used to, while saving your senior engineers valuable development hours. While someone's MentralAI likeness is done a hard day's work, it update's its human on what was accomplished so that they can intervene and get involved where they deem necessary. MentralAI is targeted at corporations who having invested heavily in their senior engineers, want to make the most of their investment. MentralAI is a solution that allows corporations to empower their employees with a tool that can save them valuable time, increasing productivity and ROI on the companys hires.
What it does
Mentral AI creates a version of your peer, mentor or role models by analyzing their GitHub profiles. These AI mentors replicate how your actual coworkers think, any specific code philosophies, and communication patterns. Chat with them anytime through voice or suggestions for code reviews and architectural guidelines.
How we built it
It is a VS Code extension in TypeScript. Behind the scenes, we use a GitHub analysis engine to extract personality traits from repos, commits, and coding habits. We integrated that with Google Gemini with custom prompts with a RAG like pipeline combining AST(Abstract Syntax Trees) insights and GitHub derived traits while VAPI powers voice conversations.
Challenges we ran into
Integrating many different API's wasn't straightforward and sometimes we had to make workarounds to solve VS Code's specific limitations. VS Code does not have a proper built-in microphone capabilities so we had to develop a work-around! Also, making sure the personalities are authentic was another challenge as translating coding style into AI prompts took a lot of iteration.
Accomplishments that we're proud of
We are proud that we created real mentors that inspire based off analyzing Github profiles. These AI mentors don't just give generic advice, but give specific style, tone and judgement of actual teammates.
What we learned
Building AI that truly reflects a team’s style is as much about understanding human nuance as it is about technology. We learned that small details in commits, coding habits, and phrasing make a big difference in creating authentic, helpful mentors.
What's next for Mentral AI
We want to expand MentralAI to be able to create a full team companion that can take multiple perspectives and have live PR reviews with checklists based on the personalities as well as expansion into different IDEs like JetBrains.
Built With
- gemini
- javascript
- nodemailer
- openai
- typescript
- vapi

Log in or sign up for Devpost to join the conversation.