mykin-ai / ml-coding-challenge-2024

Take home challenge

Technical Challenge for Machine Learning Engineer at Kin

About Kin

At Kin, we envision Personal AI as the most significant digital asset everyone will own. It should not be confined within walled gardens but should transcend ecosystems and be ultimately owned and controlled by its user.

Kin serves as your AI companion, offering personalized support with long-term memory and a privacy-by-design framework. With each conversation, Kin learns more about your world, guiding you through situations over time and providing insights into your past interactions.

The Role

We are seeking a Machine Learning Engineer to enhance the conversational AI capabilities of Kin. This role is pivotal in building and refining the technology that enables Kin to function as a truly personalized AI companion. You'll be working with large language models and conversational AI to ensure that interactions with Kin are as natural and effective as possible.

Responsibilities

• Transition from closed-source LLM APIs to open-source LLMs.
• Optimize, evaluate, deploy, and monitor performance.
• Implement an advanced ML/LLM Ops workflow focusing on streamlined models for edge computing and improved software engineering productivity.
• Develop and optimize conversational AI models, with a focus on large language models.
• Lead initiatives to enhance the performance of language models in our primary offerings.
• Stay updated and communicate the latest developments in AI.
• Collaborate with our diverse team to seamlessly integrate AI technology into Kin's platform, advancing AI developments for an improved user experience.

Requirements

• Commitment to Kin's ethos and principles.
• Collaborative and proactive work ethic.
• Proficiency in NLP and experience with PyTorch, TensorFlow, HuggingFace, transformers, and Python.
• Excellent command of English and the ability to clearly communicate technical details and overall product goals.

Challenge Overview

Duration: 2-4 hours. Please do not spend more time on this. We will have time to dicuss improvements.

Objective: Engage with the Kin app, suggest a hypothetical improvement relevant to your position, and implement a solution using mock data and python and pseudocode.

Part 1: Dog-fooding Kin

1. Download and Set Up:

   • Download the Kin app from mykin.ai.
   • Install and set up the app on your device.

2. Interaction and Analysis:

   • Interact with Kin for about 1 hour, exploring its features and understanding how it assists users.
   • Focus particularly on how the app’s advanced memory structure utilizes user data to personalize interactions.

Part 2: Suggest and Implement an Improvement

1. Identify an Improvement:

   • Based on your interaction, identify an area of improvement that could enhance user experience or functionality.

2. Describe the Improvement:

   • Document your proposed improvement, explaining its relevance and how it aligns with Kin's goals.

3. Implementation Outline:

   • Use python as much as possible but in the inetrest of time you can also use pseudocode to outline your implementation, incorporating mock data to demonstrate how the feature works. It doesn't ahve to run, but it should clearly outline your rationale so we can discuss it on a technical level. Feel free to use an external data source or mock data to highlight the functionality of your idea.

Part 3: Documentation

• Provide a document including:
  • A summary of your app interaction.
  • A detailed description and rationale for improvement and next steps in the project.
  • Your code and mock data.

Part 4: Making your code available

• Share your final work with us via git repo link, at the latest 24 hours before your interview (Github usernames: @cshulby, @simonwh, and @demchuk-alex)

Evaluation Criteria

• Understanding of Kin’s functionalities.
• Creativity and relevance of the improvement.
• Clarity and practicality of the pseudocode.
• Ability to communicate technical concepts in your documentation.

Interview Follow-Up

In your technical interview, we will discuss:

• Your experience with Kin.
• Your proposed improvement and implementation details.
• Potential modifications if more time were to be available (e.g., a full project).