Hilm - A Customer Success Story by MyTea

• 

  Our homepage. It lets you focus on the most important research questions while opening door for new ones

• 

  First, you ideate on the problem solving approach. You don't chug in data firsthand to focus on the "how"

• 

  Second, you focus on the data. You can iterate with Hilm on how to derive values. Some are readily available, but many others are not

• 

  Finally, you arrive at the analysis. As always, you can always iterate with Hilm

Inspiration

I've always been interested in the "business side" of things, especially how business owner makes decisions and executes it. As I try to learn, I found that having access to critical insights and information might make the life and death of a business. That led me to this project

Hilm is an AI consultant that helps you make decision and find critical information

As Snowflake is a B2B company, I intentionally packaged my project as a mock customer success story of a client that's trying to expand his Tea brand. It is indeed a fictional company, but I try to base the business story as solid as I can through wearing my cultural lens as an Indonesian and using a very Indonesian business as my background.

I believe through seeing this as a customer success story, we can imagine how we can develop products that empower other businesses to grow with AI, rather than being eaten by AI.

What it does

Hilm thinks with you: We designed the UX to prominently nurture discussion and brainstorming, instead of only giving you one-way answers

Hilm knows how to think: Hilm speaks the consulting lingo and knows how to structure the discussion

Hilm brings you insight: We enabled it to "estimate" data through heuristics, besides retrieving facts from the internet

How we built it

There are 3 pillars which I believe makes Hilm different than your vanilla ChatGPT

User Experience: We wanted to make a UX which is collaborative and supportive for brainstorming. The decisions business owners need to make seem simple, but are actually complex due to the number of variables that it involves and the number of people that could be affected by it. Therefore one of our key design principles is Iteration. When Hilm brings you a specific section, for example in pricing strategy, you can continuously iterate on it. You can prompt it differently, you can tell the LLM to consider different perspective and cases, you can critic the LLM, etc. You don't need to move to the next section until you're satisfied with the result, you can just focus on a section and have an enriching "debate" which will hopefully improve your decision making.

Consulting Framework: Vanilla ChatGPT does not speak the consulting lingo. Therefore we feed Hilm with specific consulting cases that mostly mimic the research question that the user is currently looking for. We use RAG to feed the most relevant business case for Hilm to reference to when solving the user's research question.

Information Gathering: We gathered data from the internet -- In this case, related to the tea industry -- We ensured the data was clean and contains insightful information. We enabled Hilm to RAG to the data, and we also equipped it with the ability to let the user fill in the data when the user feels odd with the answer. We also intentionally let it "hallucinate" in order for it to be able to heuristically derive data if it's not explicitly mentioned in the dataset. This is akin to the consulting process where users need to make heuristics to solve a problem in the first step.

Challenges we ran into

I spent most of the time on ideation

• Ideating how to make the model to be able to speak "consulting lingo", and how to translate that to RAG
• Ideating how to let the model get information from the internet, and also how to translate that to RAG
• Ideating what's the best user experience which enables conversation and coworking instead of only sending and receiving messages as we're used to with the AI chatbots, and how to translate that into Streamlit.

Accomplishments that we're proud of

Although small, I'm happy to take a step into reimagining how future decision making software would be built

What we learned

Product ideation is hard - Especially in the AI field where there's no established industry standard yet. But there's no other way but to take steps and continue iterate. And the fact that we are working with a fundamentally different paradigm of technology means that there are doors open to reinvent most products we're familiar with.

What's next for Hilm - A Customer Success Story by MyTea

Hopefully winning the prize!

Built With

• mistral
• python
• snowflake
• streamlit