How To ANYTHING

Home page
Query: "How to play Minecraft"

Inspiration

We recognized the overwhelming amount of disparate information scattered across the web. This made it hard for individuals to find concise, reliable instructions for a wide array of tasks. "How To ANYTHING" was born out of the desire to streamline this process, turning any query into a detailed guide.

What it does

"How To ANYTHING" transforms user queries into step-by-step guides akin to WikiHow articles. By leveraging advanced text and image generation techniques, our platform creates uniquely tailored instructions with associated images, ensuring that even the most intricate tasks become straightforward.

How we built it

The foundation of our platform lies in the integration of state-of-the-art text generation and image synthesis technologies. We utilize LLMs for generating detailed, accurate textual content, while sophisticated algorithms create corresponding images that enhance comprehension. We used Together AI, Aleo, zillisDB, and HUME AI The system is built on the Next.js / FastAPI framework, ensuring a seamless and responsive user experience.

Challenges we ran into

Token Limitations: When we initially built our "How to Anything" tool, we faced significant challenges due to token limitations in our text-to-image model. Many complex tasks and detailed instructions required a large number of tokens to convey accurately. This limitation hampered our ability to generate comprehensive and detailed step-by-step instructions, which was a roadblock for providing valuable user experiences.

Transition to Together AI: To address the token limitations and other challenges, we made the strategic decision to transition to Together AI. While this transition was beneficial in many ways, it also presented its own set of challenges, such as adapting our existing infrastructure and workflows to a new platform.

Accomplishments that we're proud of

Achieving a seamless blend of generated text and images that genuinely aid users in understanding complex tasks is our hallmark. The intuitive interface and the reliability of our guides have garnered positive early feedback, reinforcing our belief in the platform's potential.

What we learned

The project was a deep dive into the intricacies of text and image generation technologies. We learned about different LLMs and Stable Diffusion models. We learned how to deploy a backend and a frontend. We learned about prompt engineering, tokenization and token count, and much more!

What's next for How To ANYTHING

Our vision is to continually refine and expand the range of topics available. We're exploring the integration of videos and interactive elements, and we're also looking into fostering a community where users can contribute, modify, and rate guides, making "How To ANYTHING" a collaborative knowledge hub powered by Generative AI.