Missed the Win—Could My App Still Be a Useful? I Need Your Honest Feedback!

Congratulations to all the winners! While it’s disappointing that my app wasn’t selected, I’d love to hear feedback on my project to understand its strengths and areas for improvement.

When I developed this app, I aimed to address a gap between STEM researchers/students and industry-level tools. As an aerospace engineering student at Virginia Tech and an international student from South Korea, I’ve observed some of the challenges faced in STEM fields firsthand.

My app focuses on simplifying the use of mathematical equations and theorems, offering three powerful features tailored for STEM students and professionals:

  1. Direct Search
    Quickly find formulas or theorems by name. Get a detailed description, a rendered equation, and copyable LaTeX code, saving time compared to searching through textbooks or class notes.
  2. Multi-step Search
    For when you don’t know the exact name of a formula or theorem. Refine searches by category, field, or topic to discover what you need. Designed for students and researchers exploring new areas, the interface avoids chat-based prompts for a smoother experience.
  3. Convert
    Transform handwritten equations into professional-grade LaTeX code. Ideal for journal publications or research presentations, this feature bridges the gap between traditional note-taking and modern academic needs.

In building this app, I focused heavily on User Experience and User Interface (UI/UX), aiming to provide more accessibility than just running a system-level bash script. For the audience, I assumed it would be most useful for those with existing STEM knowledge. However, I also wanted to serve users who don’t have all the details memorized—since we’re all human and can’t retain everything we learn. This idea shaped the Multi-step Search functionality, where AI plays a critical role in handling vast amounts of information effectively.

After submitting my project, I never made a comment or post in this community. When creating my YouTube video (https://www.youtube.com/watch?v=MepsbUFQDAE), I followed advice like “show, don’t tell” and keeping it under 3 minutes as mentioned in this YouTube video. I kept the demonstration simple and focused, without voiceovers, to highlight the app’s functionality.

My GitHub repository and live website have been active since submission, but I’m unsure how the judges assessed the code or the overall usefulness of the project. I’ll admit I didn’t thoroughly research whether similar tools already existed, but my goal was to create something unique, avoiding replication of existing tools, while catering to a niche audience.

I understand my app isn’t perfect and has its limitations, but I believe it holds potential value. That’s why I’d greatly appreciate your honest feedback, even now that the competition results have been announced.

Looking forward to your thoughts!

The UI looks very well done. But here are some possible problems/suggestions:

  1. Most LLM providers (OpenAI, Anthropic,…) already have LaTeX formatting. You can ask for the formulas straight up to these chatbots inside the chat itself. Which is much more versatile.
  2. Turning into LaTeX format can be done with a simpler CNN or some variation of a NN. There are probably (not sure of this) ocr systems for LaTeX too, since most VLM models can recognize and it doesnt differ that much from normal text (visually).

It is still a good attempt. Good luck.

Thank you for the feedback! I really appreciate your insights.

When I developed this app during the competition, providers like OpenAI and Anthropic weren’t as advanced in offering LaTeX-related features as they are now. I chose not to rely on chatbot integrations because my primary goal was to provide a different kind of user experience—one that didn’t require users to manually craft prompts or have prior knowledge of AI tools like CNNs or neural networks.

The main focus of my app was to increase accessibility for users who may not be familiar with AI technologies but still need efficient tools to handle formulas and theorems. I wanted the process to feel intuitive and straightforward, without the learning curve of interacting with chatbots or other AI systems.

That said, your feedback about existing OCR systems and model capabilities is very helpful, and I’ll definitely consider these points as I think about future improvements. Thanks again for taking the time to share your thoughts!

Okay. I think I couldn’t explain myself clearly:

  • Bing chat apparently has been supporting LaTeX since 2023 (Bing Chat gets LaTeX support for mathematical equations - gHacks Tech News) and chatgpt prolly got it afterwards. And this is just for displaying, these models were already able to write it before that. But it is correct that anthropic got it few months ago, I dont remember exactly.
  • The user doesn’t need to have prior knowledge of CNNs or NNs anyways, I am telling that you don’t need Google’s LLM for this task. It is too big of a model and that you can solve a part of what your app does with a smaller and a different type of model.
1 Like

While platforms like Bing Chat and ChatGPT have integrated LaTeX support, my goal was to create a tool that offers a more intuitive and accessible experience for users who may not be familiar with AI technologies or chat-based interfaces. By providing features like Direct Search and Multi-step Search, I aimed to simplify the process of finding and converting mathematical equations without the need for crafting prompts or understanding complex models.

Regarding the use of large language models (LLMs), I understand your point that smaller, more focused models could handle specific tasks. However, my intention was to leverage the capabilities of the Gemini API to provide accurate and efficient results, especially in STEM fields where fact-based, non-hallucinated answers are critical. I believed that integrating Gemini could offer a more comprehensive and reliable solution for these needs.

I also acknowledge that similar functionalities exist, and several of the winning applications share similarities with existing tools or projects. For instance:

  • Jayu is comparable to Leon (Leon) and NeonAI (NeonAI).
  • Prospera is similar to Chorus.ai (Chorus.ai).
  • Outdraw AI resembles Quick, Draw! (Quick, Draw!).
  • Trippy shares functionality with Google Travel and Wanderlust (Wanderlust).
  • ViddyScribe has parallels with YouDescribe (YouDescribe).
  • PenApple relates to games like Slay the Spire and Cardinal Quest.

This isn’t to say the winners copied existing projects, but rather that they reimagined or built upon existing ideas in innovative ways. Similarly, my app sought to address a niche audience by focusing on accessibility and usability, particularly for users who might not find chatbot-based solutions intuitive or user-friendly.

If we avoid building applications simply because similar functionalities exist, it would limit creativity and the exploration of how tools like Gemini can be used to enhance existing ideas or solve overlooked problems.

1 Like

Redundancy is perfectly fine. Even with a completely dominant tool like Google and Wikipedia, people will use alternative ones.

Competitions are also terrible indicators of how well it would do on the market. I’ve been in startup groups where there’s always a few hackathon champions, and they usually end up doing the worst on the market… hammer looking for a nail. Startups start with the problem, before the solution.

If you’re building something for yourself, try seeing if people around you would use it. Would they pay for it? If they wouldn’t, then it’s not bringing in enough value. Maybe they don’t even pay for ChatGPT, but if it’s valuable and the entry barrier isn’t low, people would at least pay something.

I do some math work and do look up equations and write LaTeX, but I don’t really see the need for a better tool. Past a certain skill level, it’s faster to type it than copy it. I’d say something that solves equations might be better. Or something that explains formulas visually. I think the hard path with math is understanding the why what how, and none of the existing tools do that well.

Don’t let yourself discouraged. I’m pretty sure you learned a lot during developing it. Latex and the equations are beautiful things.

1 Like

my project was not selected too but I have started converting it into a SaaS, and I foresee it to be able to make me $20,000 to $50,000 monthly if it succeeds

So you could do the same too, the fact that Google didn’t select it doesn’t make it a bad project, and based on the description you provided it’s really a great project that can do well as a SaaS

I am the creator of Trippy, while my reply may seem a bit defensive, it is not my intention but seeing you compared it to Google Flights and Wonderlust, I will like to talk about Trippy, first by going through what it takes to plan a trip using Trippy.

Firstly, I really dislike forms. Don’t get me wrong, forms have helped humans a lot to be able to capture structured data but in the context of modern technology, I believe AI has gotten good enough that in some situations we don’t need to have all these forms. This was the first thing I wanted to do with Trippy. All travel platforms will present users with some sort of form to fill, it could be your origin and destination, travel dates, entourage etc.

On Trippy, a simple 5 second recording or sentence is enough.
When you open the app, this is the first screen you see. Based on your current (GPS) location, or the location of your origin country, you get to see a list of all the countries you can visit be it visa free, visa on arrival etc etc.

Now, when planning trips with Trippy, you can record a short voice message, the same way you do on Whatsapp or you can enter a short text description. For the purpose of this I will be using the privary mode which lets you enter a description rather than recording.

In the description , it’s very simple, but really the possibilities are endless. You can describe a trip involving your family members (both adults and children), involving multiple cities, involving different types of flights (business, economy etc), add dietary information, accomodation preferences etc.

The first thing on your trip plan is a summary page, which will display all your destinations, your least budget calculation and more. This is done in your currency. So somene using dollars will see this in dollars (currency is automatically converted), FCFA, pounds, naira etc.

If you open any of the countries in your destination, there’s structured information summarized for everything about that country from culture, systems of government and laws, basically going through this section can already give you a strong feeling of knowing the country even if you have never been there before.

Next is flights, this is actually real time flight information. You can open each flight to see more details about it, carbon footprint etc. This is also organized into the cheapest flights and the fastest flights.

Under accomodation, this is also real time information, separated into three price categories. Opening each accomodation will give you more details about it including a summary of everything, personalized for you based on your history and also the particular trip in question. This is helpful especially if you don’t have the time or energy to read hundreds of reviews for each property. For instance if I was traveling with a child, the AI summary can already tell me if parking is nearby, convenience shops, if the accomodation has stairs or an elevator, meals, security etc. All of this is gotten from all reviews, plus property description, amenities etc.

Then we have events. Trippy shows you real time information about all the events that are taking place in each of your destinations. This is great if you have some time to kill and also wanted to network or learn something while traveling. It also provides a summary of each event which is again personalized based on your history and the particular trip in question.

Then we have the expense section. Here, you just take photos of the various receipts id your purchases. Trippy logs this for you with automatic currency conversion. This feature is good for people who love expense tracking and are keen to find out how much they spent while on their trip or vacation.

All of the above information as you can imagine utilizes different specialized APIs and if you’ve ever tried to build a travel application, you’ll soon realize having access to APIs is not so easy unless you are already a big flight company. However, Trippy is able to get all this real time information with some smart coding tricks, I won’t go into much detail about the technical aspect of how this is actually done.

Again, Trippy can do all of this in your language, so you could write your trip plan in German, or your audio in chinese and Trippy will always respond to you in that language. I think this is another great feature of Gemini AI in that we can build truly multilingual apps with dynamic content. Most often apps have static translation which means most things are already translated within the app and a language file is loaded based on the users preference, with AI, even the dynamic content can also be multilingual and this is something I think Gemini AI helps Trippy achieve. If my mom couldn’t speak english and knows say italian, or arabic, Trippy is perfectly fine for her to use.

It’s not uncommon for people to seek out travel agents when traveling. The role of the travel agent is to understand your travel plans and destination in a way that they can be able to answer your questions and also suggest you the best travel itenary. Trippy also has a chat feature for each trip. Think of it like your own personal travel agent who has information about the entire trip, from your plans to flights, accomodation, events, destination etc. This personal assistant can answer any and all questions you may have with real time information and in your language of choice.

Now, this is just the trip planning part of Trippy but everything is based on Firebase, authentication, database, firestore, real time notifications and more. Quite frankly it is indeed a Firebase powerhouse.

But looking back at all the information we have gotten, you can see that the entire trip is organized based on a sentence of 7 words. To me this is very impressive for Gemini to be able to manage the bulk of this data.

I really enjoyed the time building Trippy, it’s currently not released but I can say I’ve planned over a dozen trips already with it. I love using it and compared to how I previously did trip planning (I had a new browser tap for everything, multiple tabs for flights, accomodations, currency conversion etc) and I usually spent hours just going through all the data from different platforms, often getting overwhelmed just trying to plan any trip. I believe technology can be really useful especially if it saves us time, time we can spend with our loved ones or doing other things which are improve our quality of life. Trippy is a tool that saves you a ton of time. You could literally be doing a morning jog and plan an entire trip with Trippy etc. Given that about 1.4 billion people travel annually, if we were to factor in all the time taken by a quarter of these travelers in trip planning, we can certainly agree that a tool that saves time is certainly good for humanity.

A more technical article has been written by Lloyd on the Firebase blog which I am happy to share here:

https://firebase.blog/posts/2024/11/gemini-competition-best-firebase-app

I can think of a dozen new things I can add to Trippy and which I certainly will with time.

While there is more to Trippy, I hope my humble attempt above can at least give an idea of why I love the app I built so much.

6 Likes