Kelvin is a production-grade AI workflow automation platform that enables developers to interact with real tools like GitHub and Notion using natural language.
It translates high-level intent into safe, structured, tool-aware actions, reducing repetitive engineering work.
Kelvin is fully implemented, deployed, and live, and serves as a portfolio project demonstrating real-world full-stack and backend engineering skills.
Kelvin allows authenticated users to connect external tools and perform complex actions through AI-driven workflows.
Unlike simple chatbots, Kelvin:
- Operates on real APIs
- Executes multi-step actions
- Persists history and state
- Enforces authentication and permissions
- Google OAuth
- Email-based authentication
- Per-user session isolation
- Secure token handling
Kelvin can perform real GitHub operations, including:
- Create GitHub issues from natural language
- Summarise pull requests
- Edit existing files and open pull requests
- Generate and create new pull requests using AI-driven changes
All actions are scoped to the user’s authorized repositories.
Kelvin supports Notion workflows such as:
- Summarising existing Notion pages
- Creating new Notion pages generated by AI
- Persisting generated content per user
- Persistent prompt and action history per user
- Structured AI responses (not raw text dumping)
- Clear separation between intent parsing and execution
- Guardrails around destructive actions
- Frontend: Next.js, React, TypeScript, Tanstack Query, Tailwind CSS, shadcn/ui
- Backend: Next.js API routes (server-heavy architecture)
- Authentication: Google OAuth, Email Auth
- AI: Gemini API
- Integrations: GitHub API, Notion API
- Database: PostgreSQL with Prisma
- Modular, domain-based backend structure
- Clear separation of controllers, services, and integrations
- Agent-style workflow orchestration instead of single-shot prompts
- Persistent data layer for users, history, and integrations
- Production deployment with environment isolation
This project prioritises correctness, safety, and maintainability over shortcuts.
Kelvin was built to explore and demonstrate:
- Building AI systems that interact with real-world tools
- Managing authentication and third-party permissions correctly
- Designing workflows that go beyond “chat with AI”
- Handling side effects safely in an AI-driven system
- Shipping and deploying a backend-heavy product
This is not a chatbot clone.
It is an integration and orchestration platform.
Kelvin is live and deployed.
Some actions are intentionally rate-limited or restricted to prevent abuse and protect connected accounts.
- AI usage is subject to Gemini API quotas
- Some advanced or destructive actions are gated
- Limits can be expanded easily with higher-tier API access
Kelvin is a personal portfolio project.
While it is fully functional and deployed, it is not positioned as a commercial SaaS product.
All actions affecting external services (GitHub, Notion) require explicit user authorization and confirmation.
Built and maintained by Yash as a real-world demonstration of full-stack and backend engineering.