HackStack

Project Info

Winner

Best Use of MindsDB

Blik.

Devpost

Inspiration

Over the last five years, we've seen the rise and the slow decline of the crypto market. It has made some people richer, and many have suffered because of it. We realized that this problem can be solved with data and machine learning - What if we can, accurately, predict forecast for crypto tokens so that the decisions are always calculated? What if we also include a chatbot to it - so that crypto is a lot less overwhelming for the users?

What it does

Blik is an app and a machine learning model, made using MindsDB, that forecasts cryptocurrency data. Not only that, but it also comes with a chatbot that you can talk to, to make calculated decisions for your. Next trades. The questions can be as simple as "How's bitcoin been this year?" to something as personal as "I want to buy a tesla worth $50,000 by the end of next year. My salary is 4000$ per month. Which currency should I invest in?" We believe that this functionality can help the users make proper, calculated decisions into what they want to invest in. And in return, get high returns for their hard-earned money!

How we built it

Our tech stack includes: Flutter for the mobile app MindsDB for the ML model + real time finetuning Cohere for AI model and NLP from user input Python backend to interact with MindsDB and CohereAI FastAPI to connect frontend and backend. Kaggle to source the datasets of historic crypto prices

Challenges we ran into

We started off using the default model training using MindsDB, however, we realized that we would need many specific things like forecasting at specific dates, with a higher horizon etc. The mentors at the MindsDB counter helped us a real lot. With their help, we were able to set up a working prototype and were getting confident about our plan. One more challenge we ran into was that the forecasts for a particular crypto would always end up spitting the same numbers, making it difficult for users to predict Then, we ended up using the NeuralTS as our engine, which was perfect. Getting the forecasts to be as accurate as possible was definitely a challenge for us, while keeping it performant enough. Solving every small issue would give rise to another one; but thanks to the mentors and the amazing documentations, we were able to figure out the MindsDB part. Then, we were trying to implement the AI chat feature, using CohereAI. We had a great experience with the API as it was easy to use, and the chat completions were also really good. We wanted the generated data from Cohere to generate an SQL query to use on MindsDB. Getting this right was challenging, as I'd always need the same datatype in a structured format in order to be able to stitch an SQL command. We figured this also out using advanced prompting techniques and changing the way we pass the data into the SQL. We also used some code to clean up the generated text and make sure that its always compatible.

Accomplishments we're proud of

Honestly, going from an early ideation phase to an entire product in just two days, for an indie team of two college freshmen is really a moment of pride. We created a fully working product with an AI chatbot, etc. Even though we were both new to all of this - integrating crypto with AI techologies is a challenging problem, and thankfully MindsDB was very fun to work with. We are extremely happy about the mindsDB learnings as we can now implement it in our other projects to enhance them with machine learning.

What we learned

We learnt AI and machine learning, using MindsDB, interacting with AI and advanced prompting, understanding user's needs, designing beautiful apps and presenting data in a useful yet beautiful way in the app.

What's next

At Blik, long term, we plan on expanding this to a full fledged crypto trading solution, where users can sign up and create automations that they can run, to "get rich quick". Short term, we plan to increase the model's accuracy by aggregating news into it, along with the cryptocurrency information like the founder information and the market ownership of the currency. All this data can help us further develop the model to be more accurate and helpful.

Analysis

Compare with all teams

View

Metric

No commits on this project resolved to a GitHub account.

Technology

Found in codeClaimed only
  • CIn code
  • DartIn code
  • KotlinIn code
  • SwiftIn code
  • CohereClaimed
  • FastAPIClaimed
  • PythonClaimed

4 of 7 appear in the indexed code. 3 claimed on Devpost could not be matched to code, which may simply mean the tool leaves no trace in the repository.

AI coding agents

No AI coding agent signals were found in this repository.

Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.

Codebase size

Source size

82 KB

Source files

31

Counts recognized source files only; vendored directories, binaries and lockfiles are excluded, so this is smaller than the repository on disk.

0 stars