Admin Snap 🧠 About the Project Salesforce is one of the most powerful CRM platforms in the world, but it’s also complex, click-heavy, and overwhelming for many users. We wanted to change that.
This project was inspired by a simple idea:
What if anyone could control Salesforce just by typing what they wanted—like asking an assistant? Admin Snap lets users connect their Salesforce org and then use natural language to do anything—create leads, update records, run queries, delete accounts—just by prompting.
💡 What I Learned How to connect Salesforce OAuth securely using a custom Connected App
How to convert natural language into structured API calls using GPT-4
How to manage token-based access per user securely with Supabase
How to debug Salesforce errors, handle scope mismatches, and deal with token refresh strategies
🛠️ How I Built It Created a modern Next.js app with Supabase authentication
Implemented a full OAuth 2.0 flow with Salesforce and securely stored user tokens
Used OpenAI's GPT-4 to parse prompts like:
“Create a contact named Sarah Connor with email sarah@skynet.com”
“List all opportunities created this month”
Parsed AI output and forwarded it to the correct Salesforce REST API endpoints
Returned results to the dashboard for real-time user feedback
⚠️ Challenges I Faced OAuth errors from mismatched callback URLs and invalid scopes
Salesforce sandbox domain quirks and login redirects
Handling HTML error responses instead of JSON in API flows
Getting GPT-4 to reliably output clean JSON API call structures
Ensuring secure multi-user token handling with Supabase and JWT sessions
Built With
- gpt-4
- next.js
- node-fetch
- openai
- postgresql
- salesforce
- supabase
- tailwind-css
- typescript
Log in or sign up for Devpost to join the conversation.