AI-First in Action: Three Recent Instances

The primary theme of this newsletter is to speak of how AI will not only make us much more productive at work but also how it will change the nature of work itself. We have spoken about how we believe an AI-first approach is best suited to realize this and start-ups are better vehicles than established enterprises to take us there. For startups like Nurish: Voice Calorie Counter , this also signifies the capability to get more things done with fewer resources. Yet, the ability to increase quantity or do something faster does that imply better quality. There is reason to believe that doing more things or doing them fast can come at the cost of quality. However, I think this is an acceptable risk for start-ups. I posit that a startup is essentially a burgeoning collection of experiments that, once proven successful, coalesce into an interlocked set of assets, systems, and processes that deliver value to all stakeholders. Generative AI and other AI technologies diminish the costs associated with these experiments, thus facilitating more innovation in a shorter timeframe.

To demonstrate the versatility and potential of AI across various domains, I would like to share a few instances from this week at Nurish. While these aren't the sole applications of AI in our company, they show the diversity of our use of AI. I look forward to eventually compiling a comprehensive list. For now, I've selected three examples that span different aspects of our work: app architecture design, financial planning, and app testing. Our intention in sharing these insights is not to proclaim our mastery over AI but to encourage you to explore and share your own experiences. We believe there's much to learn from the community and hope our journey adds some value to your endeavors.

1. Redesigning our App Architecture

Redesigning our app architecture was a pivotal moment for us as we transitioned from our prototype to a full-fledged product. Our CTO, Suresh Manian , leveraged GitHub CoPilot to navigate this crucial phase effectively.?

First, under his direction, and with a clear understanding of our app's structure and objectives, GitHub CoPilot generated a range of ideas and suggestions for crafting a new app architecture.

Then, by breaking the idea into manageable steps and engaging iteratively, Suresh employed both GitHub CoPilot and Visual Studio Code to achieve a comprehensive modular code design. He was also able to make it apply the design principles we wanted. The result was the necessary JavaScript files and a hierarchical component structure that we needed.

This innovative design now serves as the foundation for the new Nurish app, enabling us to effortlessly modify page sequences, images, and more, thanks to its modular nature.

Summary:

• Tools used: GitHub CoPilot , Visual Studio Code
• Time invested: 4-6 hours
• Time saved: 10-12 hours. We believe that the use use of AI helped us explore more design options, using more design patterns while reducing the likelihood of errors.

2. Financial Strategy

Our initial plan involves bootstrapping the startup until we achieve a viable product and user base, at which point we'll seek external capital (if you are interested, please DM me). To facilitate this, I've explored various investment vehicles like SAFE , direct equity, and convertible debt. My experience with ChatGPT and Google Gemini during this process was mixed but insightful.

• ChatGPT was instrumental in addressing my specific queries, weighing the pros and cons tailored to our situation. This significantly deepened my understanding, despite the need to sift through some irrelevant or incorrect responses.
• The tool fell short when tasked with crafting scenarios for SAFE based on our investment plans and projected valuations, leading to multiple inaccuracies. This realization prompted me to just do it myself, finding a suitable template online to complete the task.
• ChatGPT truly excelled when I sought to refine our SAFE approach. After receiving complex advice from a trusted source, I turned to ChatGPT for further clarification, which it provided adeptly and with examples, enhancing my comprehension and subsequently refining our strategy.

Navigating ChatGPT for our financial strategy proved feasible yet challenging, demanding openness to experimentation and some validation on my part. I suspect that I am getting better at being able to use the tool in a variety of ways to coax it for better results, but I can’t be sure.?

Summary:

• Tools used: ChatGPT , Google Gemini
• Time invested: 5-7 hours
• Time saved: 10-13 hours. Here time savings are a rough estimate since the the alternative might have involved seeking professional assistance if I had run into roadblocks if I had done this on my own.

3. Test Planning

At Nurish, we lack dedicated testing resources. However, as we shift from prototype to product and the product scope expands, comprehensive testing becomes critical. I turned to ChatGPT for assistance in building this. Note that I have a persona set up with ChatGPT with customer instructions where it behaves as an all-encompassing advisor with expertise in startups, products, tech, and nutrition. Now, I asked it to add mobile testing to its repertoire as well.

• I began by asking ChatGPT to delineate various necessary testing categories, a foundational step in organizing our testing framework.
• The next step involved defining the core functional tests. Here we used the screen design created by our wonderful partners at Liquidink Design . By explaining to ChatGPT the functionality at a high-level, I was able to generate a thorough list of test cases, significantly streamlining this phase.
• The most impactful aspect was determining what exactly to test. Previous informal testing of our prototype revealed inconsistencies in my approach. A comprehensive checklist for each use case, as suggested by ChatGPT, proved to be invaluable, covering many aspects I had overlooked.

Moving forward, I plan to use ChatGPT to develop test data, primarily consisting of meal records for different user personas. I'm confident in ChatGPT's (or another LLM's) ability to excel in this task, potentially surpassing my efforts. The ultimate goal is to automate these tests with Appium( chosen after a thorough tool comparison facilitated by Google Gemini). The development of test scripts using LLM assistance is the next step in this.

Summary:

• Tools used: ChatGPT , Google Gemini
• Time invested: 3-4 hours
• Time saved: 2-5 hours so far. Although this has only been applied to select functional areas, I anticipate a significant increase in efficiency as we expand this approach across the board.

Those are some of the key areas where we used AI more in everyday work at Nurish this week. How was your week? Anything new or curious that you would like to share? We are all ears!

Suresh Manian , Ranganath Krishnamani , Tvishaa Shah , Atharva Kamat