ProtocolSense

• 

  Reverse-engineer any system from examples. Powered by Gemini 3.

• 

  Pre-built examples to explore ProtocolSense instantly — pricing APIs, payment gateways, and shipping systems.

• 

  Discovered rules with confidence scores, evidence, and the exact formula — all inferred from your examples.

• 

  Ask anything. Find edge cases, validate inputs, diagnose failures — powered by Gemini 3.

• 

  Compare API versions. See what changed and what could break.

• 

  Upload request-response logs and import examples in bulk.

• 

  Paste a cURL command and extract the input-output pair automatically.

• 

  Validate protocols in your pipeline. Catch breaking changes before production.

• 

  Generate TypeScript, Python, Zod schemas, or OpenAPI specs from discovered rules. Ready to ship.

Inspiration

As a designer, I work with developers all the time. And I keep hearing the same frustration — "this API has no docs", "the guy who built it left", "I've been debugging this for two days". I've watched engineers waste hours reverse-engineering systems that should've been documented years ago.

I thought — what if there was a tool that could just figure out how something works from examples? You show it inputs and outputs, and it tells you the rules. That's ProtocolSense.

What it does

ProtocolSense takes input-output examples and reverse-engineers the hidden logic using Gemini 3.

Feed it request-response pairs — from cURL commands, CSV logs, or plain English — and it discovers the rules. Things like "books cost $20, members get 20% off, orders over $100 ship free". Then ask it "what could break?" and it finds edge cases before they become production bugs. When you're ready, export as TypeScript, Python, or OpenAPI.

How we built it

Built on Google AI Studio using Gemini 3 Flash. Frontend is React with TypeScript, styled with Tailwind CSS, deployed on Google Cloud Run. AI Studio made prototyping fast — test prompts, tweak outputs, iterate quickly.

How we use Gemini 3

Gemini 3 is the core — not a wrapper:

• Rule inference — Finds patterns from input-output pairs with confidence scores
• Edge case detection — Spots potential breaking points humans miss
• Plain English extraction — "2 books for a member costs 32" becomes structured JSON
• Failure diagnosis — Explains why inputs fail and suggests fixes
• Code generation — Exports production-ready TypeScript, Python, Zod, or OpenAPI

Challenges we faced

• Getting consistent structured JSON from Gemini required careful prompt engineering
• Balancing analysis depth with response time
• Handling API overload with retry logic

What we learned

Gemini 3 can infer real business logic from just a few examples — things like tiered discounts and conditional shipping rules. AI Studio made it possible to go from idea to working product in days.

What's next

• GitHub integration to analyze code alongside behavior
• Team collaboration features
• Webhook endpoints for CI/CD validation

Built With

• aistudio
• gemini3
• google-cloud
• materialui
• react