Inspiration
I was inspired by ChatGPT and Gemini but I quickly realised that South Africa has not been playing in the AI Space. ChatGPT is expensive for a poor kid or person like me, so I thought maybe I should build a cheaper effective version of ChatGPT. I realised as I am learning about AI that compute is very expensive and ChatGPT is being kind to be honest but it still doesn't solve the issue of affordability. As I was learning and building with AI, I saw a Firebase Studio freemium and I have been using it to build ever since and Google just made life easier and compute cheaper. I became addicted to building for free and cheaper and eventually I built milkywhay.com.
Why MILKYWHAY? Initially I wanted to use milkyway.com but the domain was expensive, so I went for milkywhay 😂. I love space, I love learning about the galaxy and exploring the unknown using the known and that is what whole AI feels like currently, it feels like anything can happen right? But there is always a question of distance, when and how quickly we reach a certain point with AI. As our slogan say "explorer beyond imagination" AI space currently is like exploring beyond your imagination and I am here to explore. Africa will have the youngest population in the world in 10 years’ time and I want our AI to creep in now by exploring the African youth needs, get them familiar with our product by year 10 we will be dominating African AI space, our products and services would have been normalised.My focus is Africa, Latin America.
This feels like a yap than what inspired me but to be honest affordability is the answer, Africa needs access to affordable resources, Google helped me with free building resources so I need to offer others cheap affordable tools too. Whoever is reading this probably taking those Google tools such as Firebase Studio and Firestore Storage for granted but I wouldn’t afford to build anything without Google seamless integration, literally. I don’t even have $2 right now as I am writing this.
What it does
MILKYWHAY is generative AI for text, image and will be introducing video too. I am also drafting a "book of heart" feature for our users to make the AI and algorithm feel like something personal, customized or of their own. Currently AI feels like a tool and everyone wants to build a tool to write emails and stuff, it's just tools everywhere. I want to build something personal for our users: a friend, consultant, a journal and a bot to brainstorm ideas, tell about your day, your thoughts and more, something like an interactive journal that you can take anywhere and feel free to say anything. It will be encrypted of course and have guardrails. People don’t need more tools, they need friends. You dont want tools in your home that is why they are kept in the garage for specific occasions. We also have deep research.
How we built it
I built the whole app on Firebase Studio, for text I am using Gemini then for images I am using Imagen. I am also using Firebase Storage, and I will be integrating Postgres for better AI memory.
Challenges we ran into
The biggest challenges I faced were API key initialisation, SDKs to be specific, those took me like two days to figure them out. Error debugging integrating Google Search on Firebase is horrible even if you manually code the whole thing. Even if you call Genkit and it just says "tool is non existent". Same with Maps and stuff, they just give you problems. Firebase needs to work on a better seamless system. JSON parsing errors are terrible to debbug, Gemini really struggles with those.
Accomplishments that we're proud of
I am glad I built a functional app at last.
What we learned
I learned that persistence is key. I learned about using NextJS. I learned building is actually the easiest step, you still need to get customers, marketing, pay for compute and fine tune, fine-tuning is expensive. Most of all, security for your app is important. I have seen people hardcoded their API keys on their code on other platforms. Through this building I learned local env and secret key managers, something I had no clue about.
What's next for MILKYWHAY
God willing, I have enough money to finetune a better image generator and afford the Veo API key to introduce a video generator. That would be enough to launch and offer monthly subscriptions. Currently our AI feels half baked for people to pay for.
We need to focus on limiting tokenization costs, we need to build better and integrate better tools. Most importantly we need to stay as lean and as effective as possible to offer cheaper services. We need to focus more on personalisation than tooling, that is what keeps retention, we want to make AI feel personal rather than a tool. Something to use every day and is part of your everyday life to assist with homework, cooking in your home, assisting with shopping not just writing emails and citations. The goal is to think beyond just a tool!!!
Built With
- firebasestudio
- germini
- tavily
Log in or sign up for Devpost to join the conversation.