I've been working on integrating AI-powered features into my e-commerce platform for the past year. Honestly, it's been a wild ride. My team and I decided to contribute to an open-source project that aligns with our goals, hoping to give back to the community and gain some experience. We spent 3 months submitting 59 pull requests, but none of them were merged. I remember last Tuesday, I was going through our 3-server setup, trying to figure out why our changes weren't being accepted.

The project we contributed to is a popular AI-powered chatbot built using Node.js and TensorFlow.js. We chose this project because it aligns with our system's requirements, and we believed that our contributions could make a significant impact. Our goal was to improve the chatbot's natural language processing capabilities by integrating a custom intent recognition model. The thing is, we didn't just want to contribute to the project - we wanted to learn from it too.

We started by forking the repository and creating a new branch for our contributions. Our first PR was a simple bug fix, which we thought would be easy to get merged. However, the maintainers asked us to add more test cases and improve the code documentation. We obliged, and here's an example of how we added test cases using Jest:

// tests/intentRecognition.test.js
const intentRecognition = require('../intentRecognition');

describe('intentRecognition', () => {
  it('should recognize intent correctly', () => {
    const input = 'I want to buy a product';
    const expectedOutput = 'buy_product';
    expect(intentRecognition(input)).toBe(expectedOutput);
  });
});

We also improved the code documentation by adding JSDoc comments:

// intentRecognition.js
/**
 * Recognizes the intent behind a given input string.
 * @param {string} input - The input string to recognize intent from.
 * @returns {string} The recognized intent.
 */
function intentRecognition(input) {
  // implementation
}

One of our PRs aimed to improve the chatbot's performance by reducing the latency of the intent recognition model. Turns out, we were able to achieve this by optimizing the model's architecture and using a more efficient algorithm. We measured the performance improvement using the performance-now library:

// performanceTest.js
const performance = require('performance-now');
const intentRecognition = require('../intentRecognition');

const input = 'I want to buy a product';
const startTime = performance();
const output = intentRecognition(input);
const endTime = performance();
const latency = endTime - startTime;
console.log(`Latency: ${latency}ms`);

Our optimization reduced the latency from 500ms to 200ms, resulting in a 60% improvement. This was a significant win for us, and we were excited to see the impact it could have on our system.

Although none of our PRs were merged, we still benefited from the experience. We learned how to optimize our system's performance, which resulted in a 30% reduction in cloud costs. We also improved our testing and documentation processes, which saved us 10 hours of development time per week. This was a huge win for us, and we were able to apply these lessons to our production system.

We spent a total of 240 hours working on the PRs, which is equivalent to 6 weeks of full-time development. Although it was a significant investment, we gained valuable experience and insights that we can apply to our system. We also developed a custom AI agent that we can use in production, which will save us 20 hours of development time per month.

In the end, I realized that the time and effort we invested were not wasted. We saved around 40 hours of development time and $1,500 in cloud costs per month. And when I did the math, I was amazed to see that we saved our company $27,000 over the next 6 months. It just goes to show that even unmerged PRs can have a significant impact on your business.